Lectures On Algebraic Groups: Alexander Kleshchev

Lectures on Algebraic Groups
Alexander Kleshchev
Contents
Part one: Algebraic Geometry page 1

1 General Algebra 3
2 Commutative Algebra 5
2.1 Some random facts 5
2.2 Ring extensions 8
3 Affine and Projective Algebraic Sets 18
3.1 Zariski topology 18
3.2 Nullstellensatz 20
3.3 Regular functions 22
3.4 Irreducible components 23
3.5 Category of algebraic sets 25
3.6 Products 28
3.7 Rational functions 29
3.8 Projective n-space 32
3.9 Functions 34
3.10 Product of projective algebraic sets 35
3.11 Example: Grassmann varieties and flag varieties 36
3.12 Example: Veronese variety 39
3.13 Problems 41
4 Varieties 46
4.1 Affine varieties 46
4.2 Prevarieties 49
4.3 Products 51
4.4 Varieties 53
4.5 Dimension 54
4.6 Problems 60
5 Morphisms 63
5.1 Fibers 63
iii
iv Contents
5.2 Finite morphisms 64
5.3 Image of a morphism 66
5.4 Open and birational morphisms 69
5.5 Problems 70
6 Tangent spaces 72
6.1 Definition of tangent space 72
6.2 Simple points 74
6.3 Local ring of a simple point 76
6.4 Differential of a morphism 77
6.5 Module of differentials 78
6.6 Simple points revisited 83
6.7 Separable morphisms 85
6.8 Problems 87
7 Complete Varieties 88
7.1 Main Properties 88
7.2 Completeness of projective varieties 89
Part two: Algebraic Groups 91
8 Basic Concepts 93
8.1 Definition and first examples 93
8.2 First properties 95
8.3 Actions of Algebraic Groups 98
8.4 Linear Algebraic Groups 100
8.5 Problems 102
9 Lie algebra of an algebraic group 105
9.1 Definitions 105
9.2 Examples 107
9.3 Ad and ad 108
9.4 Properties of subgroups and subalgebras 110
9.5 Automorphisms and derivations 111
9.6 Problems 112
10 Quotients 114
10.1 Construction 114
10.2 Quotients 115
10.3 Normal subgroups 118
10.4 Problems 120
11 Semisimple and unipotent elements 121
11.1 Jordan-Chevalley decomposition 121
11.2 Unipotent algebraic groups 124
11.3 Problems 125
Contents v
12 Characteristic 0 theory 127
12.1 Correspondence between groups and Lie algebras 127
12.2 Semisimple groups and Lie algebras 129
12.3 Problems 130
13 Semisimple Lie algebras 131
13.1 Root systems 131
13.2 Semisimple Lie algebras 134
13.3 Construction of simple Lie algebras 137
13.4 Kostant Z-form 139
13.5 Weights and representations 140
13.6 Problems 142
14 The Chevalley construction 145
14.1 Definition and first properties 146
15 Borel subgroups and flag varieties 148
15.1 Complete varieties and Borel’s fixed point theorem 148
15.2 Borel subgroups 149
15.3 The Bruhat order 151
16 The classification of reductive algebraic groups 154
16.1 Maximal tori and the root system 154
16.2 Sketch of the classification 156
Bibliography 160
Part one
Algebraic Geometry
1
General Algebra
Definition 1.0.1 A functor F : A → B is called faithful if the map
HomA (A1 , A2 ) → HomB (F (A1 ), F (A2 )), θ 7→ F (θ) (1.1)
is injective, and F is called full if the map (1.1) is surjective.
Theorem 1.0.2 A functor F : A → B is an equivalence of categories if

and only if the following two conditions hold:
(i) F is full and faithful;
(ii) every object of B is isomorphic to an object of the form F (A) for
some A ∈ Ob A.
Proof ( ⇒ ) Let F be an equivalence of categories and G : B → A

be the quasi-inverse functor. Let α : GF → idA and β : F G → idB
be isomorphisms of functors. First of all, for any object B of B βB :
F (G(B)) → B is an isomorphism, which gives (ii). Next, for each ϕ ∈
HomA (A1 , A2 ) we have the commutative diagram
αA1
GF (A1 ) - A1
GF (ϕ) ϕ
? αA2 ?
GF (A2 ) - A2
Hence ϕ can be recovered from F (ϕ) by the formula
ϕ = αA2 ◦ GF (ϕ) ◦ (αA1 )−1 . (1.2)
This shows that F is faithful. Similarly, G is faithful. To prove that F
3
4 General Algebra
is full, consider an arbitrary morphism ψ ∈ HomB (F (A1 ), F (A2 )), and
set
ϕ := αA2 ◦ G(ψ) ◦ (αA1 )−1 ∈ HomA (A1 , A2 ).
Comparing this with (1.2) and taking into account that αA1 and αA2
are isomorphisms, we deduce that G(ψ) = GF (ϕ). As G is faithful,
this implies that ψ = F (ϕ), which completes the proof that F is a full
functor.
( ⇐ ) Assume that (i) and (ii) hold. In view of (i), we can (and will)
identify the set HomB (F (A1 ), F (A2 )) with the set HomA (A1 , A2 ) for
any A1 , A2 ∈ Ob A. Using (ii), for each object B in B we can pick an
object AB in A and an isomorphism βB : F (AB ) → B. We define a
functor G : B → A which will turn out to be a quasi-inverse functor to
F . on the objects we set G(B) = AB for any B ∈ Ob B. To define G on
the morphisms, let ψ ∈ HomB (B1 , B2 ).
−1
G(ψ) := βB 2
◦ ψ ◦ βB1 ∈HomB (F G(B1 ), F G(B2 ))
= HomA (G(B1 ), G(B2 )).
It is easy to see that G is a functor, and β = {βB } : F G → idB is
an isomorphism of functors. Further, βF (A) = F (αA ) for the unique
morphism αA : GF (A) → A. Finally, it is not hard to see that α =
{αA } : GF → idA is an isomorphism of functors.
2
Commutative Algebra
Here we collect some theorems from commutative algebra which are not
always covered in 600 algebra. All rings and algebras are assumed to be
commutative.
2.1 Some random facts
Lemma 2.1.1 Let k be a field, f, g ∈ k[x, y], and assume that f is

irreducible. If g is not divisible by f , then the system f (x, y) = g(x, y) =
0 has only finitely many solutions.
Proof See [Sh, 1.1].
Proposition 2.1.2 Let A, B be k-algebras, I / A, J / B be ideals. Then
A/I ⊗k B/J → (A ⊗k B)/(A ⊗ J + I ⊗ B), ā ⊗ b̄ 7→ a ⊗ b
is an isomorphism of algebras.
Definition 2.1.3 A subset S of a commutative ring R is called multi-

plicative if 1 ∈ S and s1 s2 ∈ S whenever s1 , s2 ∈ S. A multiplicative
subset is called proper if 0 6∈ S.
Lemma 2.1.4 Let S ⊂ R be a proper multiplicative set. Let I be an

ideal of R satisfying I ∩ S = ∅. The set T of ideals J ⊇ I such that
J ∩ S = ∅ has maximal elements, and each maximal element in T is a
prime ideal.
5
6 Commutative Algebra
Proof That the set T has maximal elements follows from Zorn Lemma.
Let M be such an element. Assume that x, y ∈ R \ M . By the choice of
M , M + Rx contains some s1 ∈ S and M + Ry contains some s2 ∈ S,
i.e. s1 = m1 + r1 x and s2 = m2 + r2 y. Hence
s1 s2 = (m1 + r1 x)(m2 + r2 y) ∈ M + Rxy.
It follws that M + Rxy 6= M , i.e. xy 6∈ M .
Theorem 2.1.5 (Prime Avoidance Theorem) Let P1 , . . . , Pn be

prime ideals of the ring R. If some ideal I is contained in the union
P1 ∪ · · · ∪ PN , then I is already contained in some Pi .
Proof We can assume that none of the prime ideals is contained in

another, because then we could omit it. Fix an i0 ∈ {1, . . . , N } and for
each i 6= i0 choose an fi ∈ Pi , fi 6∈ Pi0 , and choose an fi0 ∈ I, fi0 6∈ Pi0 .
Q
Then hi0 := fi lies in each Pi with i 6= i0 and I but not in Pi0 . Now,
P
hi lies in I but not in any Pi .
Lemma 2.1.6 (Nakayama’s Lemma) Let M be a finitely generated

module over the ring A. Let I be an ideal in A such that for any a ∈ 1+I,
aM = 0 implies M = 0. Then IM = M implies M = 0.
Proof Let m1 , . . . , ml be generators of M . The condition IM = M

means that
Xl
mi = xij mj (1 ≤ i ≤ l).
j=1
for some xij ∈ I. Hence

l
X
(xij − δij )mj = 0 (1 ≤ i ≤ l).
j=1
So by Cramer’s rule, dmj = 0, where d = det(xij − δij ). Hence dM = 0.

But d ∈ 1 + I, so M = 0.
Corollary 2.1.7 If B ⊃ A is a ring extension, and B is finitely gener-

ated as an A-module, then IB 6= B for any proper ideal I of A.
Proof Since B contains 1, we have aB = 0 only if a = 0. Now all

elements of 1 + I are non-zero for a proper ideal I, so we can apply
Nakayama’s Lemma.
2.1 Some random facts 7
Corollary 2.1.8 (Nakayama’s Lemma) Let M be a finitely generated
module over the ring A, M 0 ⊆ M be a submodule, and let I be an ideal
in A such that all elements of 1 + I are invertible. Then IM + M 0 = M
implies M 0 = M .
Proof Apply Lemma 2.1.6 to M/M 0 .
Another version:
Corollary 2.1.9 (Nakayama’s Lemma) Let M be a finitely generated

module over a ring A, and I be a maximal ideal of A. If IM = M , then
there exists x 6∈ M such that xM = 0.
Proof Localize at I and apply Corollary 2.1.8.
Corollary 2.1.10 Let M be a finitely generated module over the ring A

and let I be an ideal in A such that all elements of 1 + I are invertible.
Then elements m1 , . . . , mn ∈ M generate M if and only if their images
generate M/IM .
Proof Apply Corollary 2.1.8 to M 0 = (m1 , . . . , mn ).
Lemma 2.1.11 Let M be a maximal ideal of R, then the map R →

RM induces the isomorphism of the fields R/M and RM /M RM . If we
identify the fields via this isomorphism, then the the map R → RM also
induces the isomorphism of vector spaces M/M 2 → ˜ M RM /(M RM )2 .
A field extension K/k is called separable, if either char k = 0 or

char k = p > 0 and for any k-linearly independent elements x1 , . . . , xn ∈
K, we have xp1 , . . . , xpn are linearly independent. A filed K = k(x1 , . . . , xn )
is called separably generated over k if K is a finite separable extension
of a purely transcendental extension of k.
Theorem 2.1.12
(i) The extension K = k(x1 , . . . , xn )/k is separably generated if and
only if K/k is separable.
(ii) If k is perfect (in particular algebraically closed), then any field
extension K/k is separable.
(iii) Let F/K/k be field extensions. If F/k is separable, then K/k is
separable. If F/K and K/k are separable, then F/k is separable.
Theorem 2.1.13 (Primitive Element Theorem) If K/k is a finite
separable extension, then there is an element x ∈ K such that K = k(x).
Let L/E be a field extension. A derivation is a map δ : E → L such

that
δ(x + y) = δ(x) + δ(y) and δ(xy) = xδ(y) + δ(x)y (x, y ∈ E).
If F is a subfield of E, then the derivation δ is F -derivation if it is F -

linear. The space DerF (E, L) of all F -derivations is a vector space over
L. With this notation, we have:
Theorem 2.1.14
(i) If E/F is separably generated then
dim DerF (E, L) = tr. degF E.
(ii) E/F is separable if and only all derivations F → L extend to

derivations E → L.
(iii) If char E = p > 0, then all derivations are zero on the subfield
E p . In particular, if E is perfect, all derivations of E are zero.
Theorem 2.1.15 [Ma, Theorem 20.3] A regular local ring is a UFD. In

particular it is an integrally closed domain.
2.2 Ring extensions
Definition 2.2.1 A ring extension of a ring R is a ring A of which R is

a subring.
If A is a ring extension of R, A is a fathful R-module in a natural way.

Let A be a ring extension of R and S be a subset of A. The subring
of A generated by R and S is denoted R[S]. It is quite clear that R[S]
consists of all R-linear combinations of products of elements of S.
Definition 2.2.2 A ring extension A of R is called finitely generated if

A = R[s1 , . . . , sn ] for some finitely many elements s1 , . . . , sn ∈ A.
The following notion resembles that of an algebraic element for field

extensions.
Definition 2.2.3 Let A be a ring extension of R. An element α ∈ A is
called integral over R if f (α) = 0 for some monic polynomial f (x) ∈ R[x].
A ring extension R ⊆ A is called integral if every element of A is integral
over R.
In Proposition 2.2.5 we give two equivalent reformulations of the in-

tegrality condition. For the proof we will need the following technical
Lemma 2.2.4 Let V be an R-module. Assume that v1 , . . . , vn ∈ V and

Pn
aij ∈ R, 1 ≤ i, j ≤ n satisfy j=1 akj vj = 0 for all 1 ≤ k ≤ n. Then
D := det(aij ) satisfies Dvi = 0 for all 1 ≤ i ≤ n.
Pn
Proof We expand D by the ith column to get D = k=1 aki Cki , where
Pn
Cki is the (k, i) cofactor. We then also have k=1 akj Cki = 0 for i 6= j.
So
n
X n
X XX n
Dvi = aki Cki vi = aki Cki vi + ( akj Cki )vj
k=1 k=1 j6=i k=1
n X
X n n
X n
X
= akj Cki vj = Cki akj vj = 0.
j=1 k=1 k=1 j=1
Proposition 2.2.5 Let A be a ring extension of R and α ∈ A. The

following conditions are equivalent:
(i) α is integral over R.
(ii) R[α] is a finitely generated R-module.
(iii) There exists a faithful R[α]-module which is finitely generated as
an R-module.
Proof (i) ⇒ (ii) Assume f (α) = 0 , where f (x) ∈ R[x] is monic of

degree n. Let β ∈ R[α]. Then β = g(α) for some g ∈ R[x]. As f is
monic, we can write g = f q +r, where deg r < n. Then β = g(α) = r(α).
Thus R[α] is generated by 1, α, . . . , αn−1 as an R-module.
(ii) ⇒ (iii) is clear.
(iii) ⇒ (i) Let V be a faithful R[α]-module which is generated as an
R-module by finitely many elements v1 , . . . , vn . Write
αvi = ai1 v1 + · · · + ain vn (1 ≤ i ≤ n).

Then
−ai1 v1 − · · · − ai,i−1 vi−1 + (α − aii )vi − ai,i+1 vi+1 − · · · − ain vn = 0
for all 1 ≤ i ≤ n. By Lemma 2.2.4, we have Dvi = 0 for all i, where

α − a11 −a12 ··· −a1n

−a21 α − a22 · · · −a2n
D= . .

.. .. .. ..
. . .

−a an2 · · · α − ann
n1
As v1 , . . . , vn generate V , this implies that D annihilates V . As V is

faithful, D = 0. Expanding D shows that D = f (α) for some monic
polynomial f (x) ∈ R[x].
Lemma 2.2.6 Let R ⊆ A ⊆ B be ring extensions. If A is finitely

generated as an R-module and B is finitely generated as an A-module,
then B is finitely generated as an R-module.
Proof If a1 , . . . , am are generators of the R-module A and b1 , . . . , bn

are generators of the A-module B, then it is easy to see that {ai bj } are
generators of the R-module B.
Proposition 2.2.7 Let A be a ring extension of R.

(i) If A is finitely generated as an R-module, then A is integral over
R.
(ii) If A = R[α1 , . . . , αn ] and α1 , . . . , αn are integral over R, then A
is finitely generated as an R-module and hence integral over R.
(iii) If A = R[S] and every s ∈ S is integral over R, then A is integral
over R.
Proof (i) Let α ∈ A. Then A is a faithful R[α]-module, and we can

apply Proposition 2.2.5.
(ii) Note that R[α1 , . . . , αi ] = R[α1 , . . . , αi−1 ][αi ]. Now apply induc-
tion, Proposition 2.2.5 and Lemma 2.2.6.
(iii) Follows from (ii).
Corollary 2.2.8 Let A be a ring extension of R. The elements of A

which are integral over R form a subring of A.
Proof If α1 , α2 ∈ A are integral, then α1 − α2 and α1 α2 belong to

R[α1 , α2 ]. So we can apply Proposition 2.2.7(ii).
This result allows us to give the following definition
Definition 2.2.9 The integral closure of R in A ⊇ R is the ring R̄ of all

elements of A that are integral over R. The ring R is integrally closed
in A ⊇ R in case R̄ = R. A domain R is called integrally closed if it is
integrally closed in its field of fractions.
Example 2.2.10 The elements of the integral closure of Z in C are

called algebraic integers. They form a subring of C. In fact the field of
algebraic numbers A is the quotient field of this ring.
We record some further nice properties of integral extensions.
Proposition 2.2.11 Let R, A, B be rings.

(i) If R ⊆ A ⊆ B then B is integral over R if and only if B is
integral over A and A is integral over R.
(ii) If B is integral over A and R[B] makes sense then R[B] is integral
over R[A].
(iii) If A is integral over R and ϕ : A → B is a ring homomorphism
then ϕ(A) is integral over ϕ(R).
(iv) If A is integral over R, then S −1 A is integral over S −1 R for every
proper multiplicative subset S of R.
Proof (i)-(iii) is an exercise.

(iv) First of all, it follows from definitions that S −1 R is indeed a
subring of S −1 A. Now, let [ as ] ∈ S −1 A. As [ as ] = [ a1 ][ 1s ], it suffices to
show that both [ a1 ] and [ 1s ] are integral over S −1 R. But 1s ∈ S −1 R and
for [ a1 ] we can use the monic polynomial which annihilates a.
It follows from Proposition 2.2.11(i) that the closure of R̄ in A ⊇ R is
again R̄. In particular, if D is any domain and F is its field of fractions,
then the closure D̄ in F is an integrally closed domain (since the quotient
field of D̄ is also F ).
We recall that a domain R is called a unique factorization domain or
UFD if every non-zero non-unit element of R can be written as a product
of irreducible elements, which is unique up to a permutation and units.
Proposition 2.2.12 Every U F D is integrally closed.
Proof Let R be a UFD and F be its field of fractions. Let ab ∈ F be

integral over R. We may assume that no irreducible element of R divides
both a and b. There is a monic polynomial f (x) = xn +rn−1 xn−1 +· · ·+
r0 ∈ R[x] with f ( ab ) = 0, which implies an + rn−1 an−1 b + · · · + r0 bn = 0.
So, if p ∈ R is an irreducible element dividing b then p divides an , and
hence p divides a, a contradiction. Therefore b is a unit and ab ∈ R.
Proposition 2.2.13 If a domain R is integrally closed, then so is S −1 R

for any proper multiplicative subset S of R.
Proof Exercise.
Example 2.2.14 The ring Z[i] of Gaussian integers is Euclidean (the

degree function is ∂(a + bi) = a2 + b2 , hence it is a UFD, and so it is
integrally closed by Proposition 2.2.13. On the other hand consider the
ring Z[2i]. The quotient field of both Z[i] and Z[2i] is Q(i), and we have
Z[2i] ⊂ Z[i] ⊂ Q[i]. Clearly Z[2i] is not integrally closed, as i 6∈ Z[2i] is
integral over it. It is easy to see that Z[2i] = Z[i].
Theorem 2.2.15 If R is integrally closed, then so is R[x1 , . . . , xr ].
Next we are going to address the question of how prime ideals of R

and A are related if A ⊇ R is an integral extension.
Definition 2.2.16 Let R ⊆ A be a ring extension. We say that a prime

ideal P of A lies over a prime ideal p of R if P ∩ R = p.
The following lemma is a key technical trick.
Lemma 2.2.17 Let A ⊇ R be an integral ring extension, p be a prime

ideal of R, and S := R \ p.
(i) Let I be an ideal of A avoiding S, and P be an ideal of A maximal
among the ideals of A which contain I and avoid S. Then P is
a prime ideal of A lying over p.
(ii) If P is a prime ideal of A which lies over p, then P is maximal
in the set T of all ideals in A which avoid S.
Proof (i) Clearly, S is a proper multiplicative subset of A. So P is prime

in view of Lemma 2.1.4. We claim that P ∩ R = p. That P ∩ R ⊆ p is
clear as P ∩ S = ∅.
Assume that P ∩ R ( p. Let c ∈ p \ P . By the maximality of P ,
p + αc = s ∈ S for some p ∈ P and α ∈ A. As A is integral over R, we
have
0 = αn + rn−1 αn−1 + · · · + r0
for some r0 , . . . , rn−1 ∈ R. Multiplying by cn yields
0 = cn αn + crn−1 cn−1 αn−1 + · · · + cn r0

= (s − p)n + crn−1 (s − p)n−1 + · · · + cn r0 .
If we decompose the last expression as the sum of monomials, then the

part which does not involve any positive powers of p looks like
x := sn + crn−1 sn−1 + · · · + cn r0 .
It follows that x ∈ P . On the other hand, x ∈ R, so x ∈ R ∩ P ⊆ p.

Now c ∈ p implies sn ∈ p. As p is prime, s ∈ p, a contradiction.
(ii) If P is not maximal in T , then there exists an ideal I in T which
properly contains P . As I still avoids S, it also lies over p. Take u ∈ I\P .
Then u 6∈ R and u is integral over R. So the set of all polynomials
f ∈ R[x] such that deg f ≥ 1 and f (u) ∈ P is non-empty. Take such
Pn
f (x) = i=0 ri xi of minimal possible degree. We have
un + rn−1 un−1 + · · · + r0 ∈ P ⊆ I,
whence r0 ∈ R ∩ I = p = R ∩ P ⊆ P . Therefore
un + rn−1 un−1 + · · · + r1 u = u(un−1 + rn−1 un−2 + · · · + r1 ) ∈ P.
By the choice of u and minimality of deg f , u 6∈ P and un−1 +rn−1 un−2 +

· · · + r1 6∈ P . We have contradiction because P is prime.
Corollary 2.2.18 (Lying Over Theorem) If A is integral over P then

for every prime ideal p of R there exists a prime ideal P of A which lies
over p. More generally, for every ideal I of A such that I ∩ R ⊆ p there
exists a prime ideal P of A which contains I and lies over p.
Corollary 2.2.19 (Going Up Theorem) Let A ⊇ R be an integral

ring extension, and p1 ⊆ p2 be prime ideals in R. If P1 is a prime ideal
of A lying over p1 , then there exists a prime ideal P2 of A such that
P1 ⊆ P2 and P2 lies over p2 .
Proof Take p = p2 and I = P1 in Lemma 2.2.17(i).

Corollary 2.2.20 (Incomparability) Let A ⊇ R be an integral ring
extension, and P1 , P2 be prime ideals of A lying over a prime ideal p of
R. Then P1 ⊆ P2 implies P1 = P2 .
Proof Use Lemma 2.2.17(ii).

The relation between prime ideals established above has further nice
properties.
Theorem 2.2.21 (Maximality) Let A ⊇ R be an integral ring exten-

sion, and P be a prime ideal of A lying over a prime ideal p of R. Then
P is maximal if and only if p is maximal.
Proof If p is not maximal, we can find a maximal ideal m ) p. By

the Going Up Theorem, there is an ideal M of A lying over m and
containing P . It is clear that M actually containg P properly, and so P
is not maximal.
Conversely, let p be maximal in R. Let M be a maximal ideal con-
taining P . Then M ∩ R ⊇ P ∩ R = p and we cannot have M ∩ R = R,
as 1R = 1S 6∈ M . It follows that M ∩ R = p. Now M = P by Incompa-
rability Theorem.
The previous results can be used to prove some useful properties con-
cerning extensions of homomorphisms.
Lemma 2.2.22 Let A ⊇ R be an integral ring extension. If R is a field

then A ⊇ R is an algebraic field extension.
Proof Let α ∈ A be a non-zero element. Then α is algebraic over R,

hence R[α] ⊆ A is a field, and α is invertible. Hence A is a field.
Proposition 2.2.23 Let A be integral over R. Every homomorphism ϕ

of R to an algebraically closed field F can be extended to A.
Proof If R is a field, then A is an algebraic field extension of R by

Lemma 2.2.22. Now the result follows from Proposition ??.
If R is local, then ker ϕ is the maximal ideal m of R. By Lying Over
and Maximality Theorems, there is an ideal M of A lying over m. The
inclusion R → A then induces an embedding of fields R/m → A/M ,
which we use to identify R/m with a subfield of A/M . Note that the
field extension A/M ⊇ R/m is algebraic. Since ker ϕ = m, ϕ factors
through the projection R → R/m. The resulting homomorphism ϕ :
R/m → F can be extended to ψ : A/M → F by Proposition ??. Now
if π : A → A/M is the natural projection, then ψ ◦ π is the desired
extension of ϕ.
Now we consider the general case. Let p := ker ϕ, a prime ideal
in R, and S = R \ p. Then S −1 A is integral over S −1 R by Proposi-
tion 2.2.11(iv). Now S −1 R = Rp is local. By the universal property of
localizations, ϕ extends to a ring homomorphism ϕ̂ : S −1 R → F . By
the local case, ϕ̂ extends to ψ̂ : S −1 A → F , and the desired extension
ψ : A → F is obtained by composing ψ with the natural homomorphism
A → S −1 A.
Proposition 2.2.24 Every homomorphism of a field k into an alge-

braically closed field can be extended to every finitely generated ring ex-
tension of k.
Proof Let ϕ : k → F be a homomorphism to an algebraically closed

field F and R be a finitely generated ring extension of k, so that R =
k[α1 , . . . , αn ] for some α1 , . . . , αn ∈ R.
First assume that R is a field. By Proposition 2.2.23, we may as-
sume that R is not algebraic over k. Let {β1 , . . . , βt } be a (necessarily
finite) transcendence base of R over k. Each α ∈ R is algebraic over
k(β1 , . . . , βt ), i.e. satisfies a polynomial ak αk + · · · + a1 α + a0 = 0 with
coefficients ak , . . . , a0 ∈ k(β1 , . . . , βt ), ak 6= 0. Multiplying by a common
denominator yields a polynomial equation
bk α k + · · · + b1 α + b 0 = 0
with coefficients bk , . . . , b0 ∈ k[β1 , . . . , βt ], bk 6= 0. Hence α is in-

tegral over k[β1 , . . . , βt , b1k ]. Applying this to α1 , . . . , αn yields non-
zero c1 , . . . , cn ∈ k[β1 , . . . , βt ] such that α1 , . . . , αn are integral over
k[β1 , . . . , βt , c11 , . . . , c1n ]. Set c = c1 . . . cn . Then α1 , . . . , αn are inte-
gral over k[β1 , . . . , βt , 1c ], and hence R is integral over k[β1 , . . . , βt , 1c ],
see Proposition 2.2.7(ii). Let cϕ be the image of c under the homomor-
phism
k[β1 , . . . , βt ] ∼
= k[x1 , . . . , xt ] → F [x1 , . . . , xt ]
induced by ϕ. As F is infinite there exist γ1 , . . . , γt ∈ F such that

cϕ (γ1 , . . . , γt ) 6= 0. By the universal property of polynomial rings, there
exists a homomorphism ψ : k[β1 , . . . , βt ] → F which extends ϕ and sends
β1 , . . . , βt to γ1 , . . . , γt , respectively. The universal property of localiza-
tions yields an extension of ψ to ring k[β1 , . . . , βt , 1c ] = k[β1 , . . . , βt ]c .
Now Proposition 2.2.23 extends ϕ to R, which completes the case where
R is a field.
Now, let R = k[α1 , . . . , αn ] be any finitely generated ring extension
of k. Let m be a maximal ideal of R and π : R → R/m be the natural
projection. Then R/m is a field extension of π(k) ∼ = k generated by
π(α1 ), . . . , π(αn ). By the first part of the proof, every homomorphism
of π(k) into F extends to R/m. Therefore every homomorphism of
k∼= π(k) extends to R.
Let K/k be a finite field extension. Consider K as a k-vector space.

Then the map x 7→ ax is a k-linear map of this vector space. Define the
norm NK/k (a) to be the determinant of this map. Note that NK/k |K × :
K × → k × is a group homomorphism.
Lemma 2.2.25 If a = a1 , . . . , as be the roots with multiplicity of the

minimal polynomial irr (a, k) (in some extension of the field K), then
Qs
NK/k (a) = ( i=1 ai )[K:k(a)] .
Proof If 1 = v1 , v2 , . . . , vr is a basis of K over k(u), then {ai vj | 0 ≤

i < s, 1 ≤ j ≤ r} is a basis of K over k in which the matrix of the map
x 7→ ax is block diagonal with blocks all equal to the companion matrix
of irr (a, k).
Lemma 2.2.26 Let S ⊆ R be integral domains with fields of fractions

k ⊆ K, S be integrally closed, and r ∈ R be integral over S. Then
irr (r, k) ∈ S[x].
Proof Let F be an extension of K which contains all roots r = r1 , . . . , rs

of irr (r, k). Then each ri is integral over S. So the coefficients of irr (r, k),
being polynomials in the ri are also integral over S. As S is integrally
closed, the coefficients belong to S.
Corollary 2.2.27 Let S ⊆ R be domains with fields of fractions k ⊆ K

such that the field extension K/k is finite. Assume that the ring exten-
sion S ⊆ R is integral and that S is integrally closed. Then NK/k (r) ∈ S
for any r ∈ R.
Proof Apply Lemmas 2.2.25 and 2.2.26.

Lemma 2.2.28 (Noether’s Normalization Lemma) Let k be a field,
and R = k[x1 , . . . , xn ] be a domain, finitely generated over k with the
field of fractions F . If tr. degk F = d, then there exist algebraically
independent over k elements S1 , . . . , Sd ∈ R such that R is integral over
k[S1 , . . . , Sd ].
Theorem 2.2.29 (Going Down Theorem) Let S ⊆ R be an integral

ring extension and S be integrally closed. Let P1 ⊇ P2 be prime ideals
of S, and Q1 be a prime ideal of R lying over P1 . Then there exists a
prime ideal Q2 ⊆ Q1 lying over P2 .
3
Affine and Projective Algebraic Sets
3.1 Zariski topology

Algebraic geometry is the subject which studies (algebraic) varieties.
Naively, varieties are just algebraic sets.
Throughout we fix an algebraically closed ground field k. (It is much
harder to develop algebraic geometry over non-algebraically closed fields
and we will not try to do this). Denote by An the affine space k n —this
is just the set of all n-tuples of elements of k.
Definition 3.1.1 Let S ⊂ k[T1 , . . . , Tn ]. A zero of the set S is an

element (x1 , . . . , xn ) of An such that f (x1 , . . . , xn ) = 0 for all f ∈ S.
The zero set of S is the set Z(S) of all zeros of S. An algebraic set in
An (or affine algebraic set) is the zero set of some set S ⊂ k[T1 , . . . , Tn ],
in which case S is called a set of equations of the algebraic set.
Example 3.1.2 The straight line x + y − 1 = 0 and the ‘circle’ x2 + y 2 −

1 = 0 are examples of algebraic sets in C2 . More generally, algebraic
sets in C2 with a single equation are called complex algebraic curves.
Note that the curve given by the equation (x + y − 1)(x2 + y 2 − 1) = 0
is the union of the line and the ‘circle’ above. On the other hand, the
zero set of {x + y − 4, x2 + y 2 − 1} consists of two points (1, 0) and (0, 1).
Finally, two more examples: ∅ = Z(1), and C2 = Z(0).
Note that Z(S) = Z((S)), where (S) is the ideal of k[T1 , . . . , Tn ] gen-
erated by S. Therefore every algebraic set is the zero set of some ideal.
Since k[T1 , . . . , Tn ] is noetherian by Hilbert’s Basis Theorem, every al-
gebraic set is the sero set of a finite set of polynomials.
Example 3.1.3 Let us try to ‘classify’ algebraic sets in A1 and A2 .
18
3.1 Zariski topology 19
(i) Algebraic sets in A1 are A1 itself and all finite subsets (including
∅).
(ii) Let X be an algebraic set in A2 . It is given by a system of
polynomial equations: f1 (T1 , T2 ) = · · · = fm (T1 , T2 ) = 0. If all
polynomials are zero, we get X = A2 . If f1 , . . . , fm do not have a
common divisor, then our system has only finitely many solutions,
see Lemma 2.1.1. Finally, let all fi have greatest common divisor
d(T1 , T2 ). Then fi = dgi , where the polynomials gi (T1 , T2 ) do not
have a common divisor. Now, X = X1 ∪ X2 , where X1 is given
by the system g1 = · · · = gm = 0, and X2 is given by d = 0. As
above X1 is a finite (possibly empty) set of points, while X2 is
given by one non-trivial equation d = 0 (and can be thought of
as a ‘curve’ in A2 ).
Proposition 3.1.4
(i) Every intersection of algebraic sets is an algebraic set; the union

of finitely many algebraic sets is an algebraic set.
(ii) An and ∅ are algebraic sets in An .
Proof (i) Let (Xj = Z(Ij ))j∈J be a family of algebraic sets, given as
zero sets of certain ideals Ij . To see that their intersection is again an
P
algebraic set, it is enough to note that ∩j∈J Z(Ij ) = Z( j∈J Ij ). For
the union, let Z(I) and Z(J) be algebraic sets corresponding to ideals I
and J, and note that Z(I) ∪ Z(J) = Z(I ∩ J) (why?).
(ii) An = Z(0) and ∅ = Z(1).
The proposition above shows that algebraic sets in An are closed sets
of some topology. This topology is called the Zariski topology. Zariski
toplology on An also induces Zariski topology on any subset of An , in
particular algebraic set. This topology is very weird and it takes time
to get used to it. The main unintuitive thing here is that the topology
is ‘highly non-Hausdorf’—its open sets are huge. For example, we saw
above that proper closed sets in k are exactly the finite sets, and so any
two non-empty open sets intersect non-trivially.
Let f ∈ k[T1 , . . . , Tn ]. The corresponding principal open set is An \
Z(f ) = {x ∈ An | f (x) 6= 0}. It is easy to see that each open set in
An is a finite union of principal open sets, so principal open sets form a
base of Zariski topology.
20 Affine and Projective Algebraic Sets
3.2 Nullstellensatz
The most important theorem of algebraic geometry is called Hilbert’s
Nullstellensatz (or theorem on zeros). It has many equivalent reformula-
tions and many corollaries. The idea of the theorem is to relate algebraic
sets in An (geometry) and ideals in k[T1 , . . . , Tn ] (commutative algebra).
We have two obvious maps
Z : {ideals in k[T1 , . . . , Tn ]} → {algebraic sets in An }
and
I : {algebraic sets in An } → {ideals in k[T1 , . . . , Tn ]}.
We have already defined Z(J) for an ideal J in k[T1 , . . . , Tn ]. As for I,
let X be any subset of An . Then the ideal I(X) is defined to be
I(X) := {f ∈ k[T1 , . . . , Tn ] | f (x1 , . . . , xn ) = 0 for all (x1 , . . . , xn ) ∈ X}.
Lemma 3.2.1 Let X be any subset of An . Then Z(I(X)) = X̄, the

closure of X in Zariski topology. In particular, if X is an algebraic set,
then Z(I(X)) = X.
Proof We have to show that for any algebraic set Z(J) containing X we
actually have Z(I(X)) ⊆ Z(J). Well, as X ⊆ Z(J), we have I(X) ⊇ J,
which in turn implies Z(I(X)) ⊆ Z(J).
Note, however, that Z and I do not give us a one-to one correspon-
dence. For example, in A1 we have Z((T )) = Z((T 2 )) = {0}, that is
the different ideals (T ) and (T 2 ) give the same algebraic set. Also, note
that I({0}) = (T ) 6= (T 2 ). Nullstellensatz sorts out problems like this
in a very satisfactory way.
The first formulation of the Nulltellensatz is as follows (don’t forget
that k is algebraically closed, otherwise the theorem is wrong):
Theorem 3.2.2 (Hilbert’s Nullstellensatz)

√ Let J be an ideal of
k[T1 , . . . , Tn ]. Then I(Z(J)) = J.
√
Proof√ First of all, it is easy to see that J ⊆ I(Z(J)). Indeed, let
f ∈ J. Then f n ∈ J. Then f n is zero at every point of Z(J). But
this implies that f is zero at every point of Z(J), i.e. f ∈ I(Z(J)).
The converse is much deeper. Let f ∈ I(Z(J)) and assume that no
power of f belongs to J. Applying Lemma 2.1.4 to the multiplicative
set {1, f, f 2 , . . . } yields a prime ideal P containing J but not f . Let
3.2 Nullstellensatz 21
R = k[T1 , . . . , Tn ]/P and π : k[T1 , . . . , Tn ] → R be the natural pro-
jection. Then R is a domain which is generated over π(k) ∼ = k by
α1 := π(T1 ), . . . , αn := π(Tn ). We identify k and π(k), and so π can
be considered as a homomorphism of k-algebras. Under this agreement,
y := f (α1 , . . . , αn ) = π(f ) 6= 0, non-zero element of R, as f 6∈ P .
By Proposition 2.2.24, the identity isomorphism k → k can be ex-
tended to a homomorphism ψ from the subring k[α1 , . . . , αn , y1 ] of the
fraction field of R to k. Then ψ(y) 6= 0. So
f (ψ(α1 ), . . . , ψ(αn )) = ψ(f (α1 , . . . , αn )) = ψ(y) 6= 0.
On the other hand, for any g ∈ J ⊆ P we have
g(ψ(α1 ), . . . , ψ(αn )) = ψ(g(α1 , . . . , αn )) = ψ(g(π(T1 ), . . . , π(Tn )))

= ψ(π(g(T1 , . . . , Tn ))) = ψ(π(g)) = ψ(0) = 0.
Thus (ψ(α1 ), . . . , ψ(αn )) is a zero of J but not of f , i.e. f 6∈ I(Z(J)), a

contradiction.
Definition√ 3.2.3 We say that an ideal I of a commutative ring R is

radical if I = I.
The following corollary is also often called Nullstellensatz.
Corollary 3.2.4 The maps I and Z induce an order-reversing bijection

between algebraic sets in An and radical ideals in k[T1 , . . . , Tn ].
Proof Note that I(X) is always a radical ideal for any subset X ⊆ An .
Now the result follows from Theorem 3.2.2 and Lemma 3.2.1.
Corollary 3.2.5 Let J1 and J2 be two ideals of k[T1 , . . . , Tn ]. Then

√ √
Z(J1 ) = Z(J2 ) if and only if J1 = J2 .
√
Proof It is clear that Z(J) = Z( J) for any ideal J, which gives the
‘if’-part. The converse follows from Theorem 3.2.2.
Corollary 3.2.6 Every proper ideal of k[T1 , . . . , Tn ] has at least one

zero in An .
√
Proof If I = k[T1 , . . . , Tn ], then I = k[T1 , . . . , Tn ]. Now the result
follows from above.
Let x = (x1 , . . . , xn ) ∈ An . Denote I({x}) by Mx , i.e.
Mx = {f ∈ k[T1 , . . . , Tn ] | f (x1 , . . . , xn ) = 0}.
Corollary 3.2.7 The mapping x 7→ Mx is a one-to-one correspondence

between An and the maximal ideals of k[T1 , . . . , Tn ].
Proof Note that the maximal ideals are radical and apply Nullstellen-
satz.
3.3 Regular functions

n
Let X ⊆ A be an algebraic set. Every polynomial f ∈ k[T1 , . . . , Tn ]
defines a k-valued function on An and hence on X via restriction. Such
functions are called regular functions on X. The regular functions form
a k-algebra with respect to the obvious ‘point-wise operations’. The
algebra is called the coordinate algebra (or coordinate ring) of X (or
simply the algebra/ring of regular functions on X) and denoted k[X].
Clearly,
k[X] =∼ k[T1 , . . . , Tn ]/I(X).
If I is an ideal of k[X] then we write Z(I) for the set of all points x ∈ X
such that f (x) = 0 for every f ∈ I, and if Z is a subset of X we denote
by I(Z) the ideal of k[X] which consists of all functions f ∈ k[X] such
that f (z) = 0 for every z ∈ Z. Note that closed subsets of X all look
like Z(I).
Now the Nullstellensatz and the correspondence theorem for ideals
imply:
Theorem 3.3.1 (Hilbert’s Nullstellensatz) Let X be an algebraic

set.
√
(i) If J is an ideal of k[X], then I(Z(J)) = J.
(ii) The maps I and Z induce an order-reversing bijection between
closed sets in X and radical ideals in k[X].
(iii) Every proper ideal of k[X] has at least one zero in X.
(iv) The mapping x 7→ Mx = {f ∈ k[X] | f (x) = 0} is a one-to-one
correspondence between X and the maximal ideals of k[X].
Definition 3.3.2 A commutative finitely generated k-algebra without

nilpotent elements is called an affine k-algebra.
3.4 Irreducible components 23
Proposition 3.3.3
(i) Let X be an algebraic set. Then k[X] is an affine k-algebra.
(ii) Every affine k-algebra A is isomorphic to k[X] for some affine
algebraic set X.
Proof (i) clear. For (ii), if A = k[α1 , . . . , αn ] is an k-algebra generated

by α1 , . . . , αn , then by the universal property of polynomial rings, A ∼ =
k[T1 , . . . , Tn ]/I for some radical ideal I. So I = I(X) for some algebraic
set X by the Nulltellensatz.
Let f ∈ k[X]. The corresponding principal open set is
Xf := X \ Z(f ) = {x ∈ X | f (x) 6= 0}. (3.1)
Each open set in X is a finite union of principal open sets, so principal

open sets form a base of Zariski topology.
Example 3.3.4
(i) If X is a point, then k[X] = k.
(ii) If X = An , then k[X] = k[T1 , . . . , Tn ].
(iii) Let X ⊂ A2 be given by the equation T1 T2 = 1. Then k[X] is
isomorphic to the localization k[t]t ∼ = k[t, t−1 ].
3.4 Irreducible components
Definition 3.4.1 A topological space is noetherian if its open sets satisfy

the ascending chain condition.
A topological space is irreducible if it cannot be written as a union of
its two proper closed subsets.
Note that a non-empty open subset of an irreducible topological space

X is dense in X, and that any two non-empty open subsets of X intersect
non-trivially. Problem 3.13.18 contains some further important proper-
ties of irreducible spaces.
Lemma 3.4.2 An with Zariski topology is noetherian. Hence the same

is true for any subspace of An .
Proof An ascending chain of open sets corresponds to a descending

chain of closed sets, which, by the Nullstellensatz, corresponds to an
ascending chain of radical ideals of k[T1 , . . . , Tn ], which stabilizes since
k[T1 , . . . , Tn ] is noetherian.
Lemma 3.4.3 Algebraic set X ⊆ An is irreducible if and only if the

ideal I(X) is prime.
Proof If X is irreducible and f1 , f2 ∈ k[T1 , . . . , Tn ] with f1 f2 ∈ I(X),

then X ⊆ Z((f1 )) ∪ Z((f2 )), and we deduce that X ⊆ Z((f1 )) or X ⊆
Z((f2 )), i.e. f1 ∈ I(X) or f2 ∈ I(X).
Conversely, if I(X) is prime and X = X1 ∪ X2 for proper closed
subsets X1 , X2 , then there are polynomials fi ∈ I(Xi ) with fi 6∈ I(X).
But f1 f2 ∈ I(X), contradiction.
Since prime ideals are radical, Lemma 3.4.3 allows us to further re-
fine the one-to-one correspondence between radical ideals and algebraic
sets: under this correspondence prime ideals correspond to irreducible
algebraic sets. Also note that X is irreducible if and only if k[X] is a
domain. So for irreducible algebraic sets X we can form the quotient
field of k[X] is called the field of rational functions on X and denoted
k(X). In a natural way, k(X) is a field extension of k.
We now establish a general fact on noetherian topological spaces,
which in some sense reduces the study of algebraic sets to that of ir-
reducible algebraic sets.
Proposition 3.4.4 Let X be a noetherian topological space. Then X

is a finite union X = X1 ∪ · · · ∪ Xr of irreducible closed subsets. If
one assumes that Xi 6⊆ Xj for all i 6= j then the Xi are unique up to
permutation. They are called the irreducible components of X and can
be characterized as the maximal irreducible closed subsets of X.
Proof Let X be a topological noetherian space for which the first state-
ment is false. Then X is reducible, hence X = X1 ∪ X10 for proper
closed subsets X1 , X10 . Moreover, the first statement is false for at
least one of X1 , X10 . Continuing this way, we get an infinite chain
X ) X1 ) · · · ) X2 ) . . . of closed subsets, which is a contradiction, as
X is noetherian.
To show uniqueness, assume that we have two irredundant decom-
positions X = X1 ∪ · · · ∪ Xr and X = X10 ∪ · · · ∪ Xs0 . For each i,
Xi ⊆ (X10 ∩ Xi ) ∪ · · · ∪ (Xs0 ∩ Xi ), so by irreducibility of Xi we may as-
0
sume that Xi ⊆ Xσ(i) for some σ(i). For the same reason, Xj0 ⊆ Xτ (j) for
some τ (j). Now the irredundancy of the decompositions implies that σ
and τ are mutually inverse bijections between {1, . . . , r} and {1, . . . , s},
0
and Xi = Xσ(i) for all i.
3.5 Category of algebraic sets

We now define morphisms between algebraic sets. Let X ⊆ An , Y ⊆ Am
be two algebraic sets and consider a map ϕ : X → Y . Let T1 , . . . , Tn
and S1 , . . . , Sm be the coordinate functions on An and Am , respectively.
Denote Si ◦ ϕ by ϕi for all 1 ≤ i ≤ m. So that we can think of ϕ
as the m-tuple of functions ϕ = (ϕ1 , . . . , ϕm ), where ϕi : X → k, and
ϕ(x) = (ϕ1 (x), . . . , ϕm (x)) ∈ Am . The map ϕ : X → Y is called
a morphism of algebraic sets or a regular map from X to Y if each
function ϕi : X → k, 1 ≤ i ≤ n is a regular function on X. It is easy to
see that algebraic sets and regular maps form a category, in particular
a composition of regular maps is a regular map again.
Now, let ϕ : X → Y be a morphism of algebraic sets as above. This
morphism defines the ‘dual’ morphism ϕ∗ : k[Y ] → k[X] of coordinate
algebras, as follows:
ϕ∗ : k[Y ] → k[X] : f 7→ f ◦ ϕ.
It is clear that ϕ∗ is a homomorphism of k-algebras. Moreover, (ϕ◦ψ)∗ =

ψ ∗ ◦ ϕ∗ and id∗ = id, i.e. we have a contravariant functor F from
the category of algebraic sets to the category of affine k-algebras. To
reiterate: F(X) = k[X] and F(ϕ) = ϕ∗ .
Theorem 3.5.1 The functor F from the category of algebraic sets (over
k) to the category of affine k-algebras is a (contravariant) equivalence of
categories.
Proof In view of Theorem 1.0.2 (for contravariant functors) and Propo-

sition 3.3.3(ii) we just need to show that ϕ 7→ ϕ∗ establishes a one-to
one correspondence between regular maps ϕ : X → Y and algebra ho-
momorphisms k[Y ] → k[X], for arbitrary fixed algebraic sets X ⊆ An
and Y ⊆ Am . Let T1 , . . . , Tn and S1 , . . . , Sm be the coordinate functions
on An and Am , respectively.
Let α : k[Y ] → k[X] be an k-algebra homomorphism. Set sj :=
Sj |Y ∈ k[Y ], 1 ≤ j ≤ m. Then α(sj ) are regular functions on X. Define
the regular map α∗ : X → Am as follows:
α∗ := (α(s1 ), . . . , α(sm )).
We claim that in fact α∗ (X) ⊆ Y . Indeed, let x ∈ X and f =

P k1 km
k ck S1 . . . Sm ∈ I(Y ), where k stands for the m-tuple (k1 , . . . , km ).
It suffices to prove that f (α∗ (x)) = 0. Using f (s1 , . . . , sm ) = 0 and the
fact that α is an algebra homomorphism, we have
f (α∗ (x)) = f (α(s1 )(x), . . . , α(sm )(x))

X
= ck (α(s1 )(x))k1 . . . (α(sm )(x))km
k
X
= α( ck sk11 . . . skmm )(x)
k
= α(f (s1 , . . . , sm ))(x) = 0.
Now, to complete the proof of the theorem, it suffices to check that

(ϕ∗ )∗ = ϕ and (α∗ )∗ = α for any regular map ϕ : X → Y and any
k-algebra homomorphism α : k[Y ] → k[X]. Well,
(ϕ∗ )∗ = (ϕ∗ (s1 ), . . . , ϕ∗ (sm )) = (ϕ1 , . . . , ϕm ) = ϕ.
On the other hand,
((α∗ )∗ )(si ) = si ◦ α∗ = α(si )
for any 1 ≤ i ≤ m. Since the si generate k[Y ], this implies that (α∗ )∗ =
α.
Corollary 3.5.2 Two (affine) algebraic sets are isomorphic if and only
if their coordinate algebras are isomorphic.
Lemma 3.5.3 Regular maps are continuous in the Zariski topology.
Proof Let ϕ : X → Y ⊆ Am be a regular map. As the topology on

Y is induced by that on Am , it suffices to prove that any regular map
ϕ : X → Am is continuous. Let Z = Z(I) be a closed subset of Am . We
claim that ϕ−1 (Z) = Z(J) where J is the ideal of k[X] generated by
ϕ∗ (I). Well, if x ∈ Z(J), then f (ϕ(x)) = ϕ∗ (f )(x) = 0 for any f ∈ I,
so ϕ(x) ∈ Z(I), i.e. x ∈ ϕ−1 (Z). The argument is easily reversed.
Remark 3.5.4 Note that regular maps from X to Y usually do not

exhaust all continuous maps from X to Y , so the category of algebraic
sets is not a full subcategory of the category of topological spaces. For
example, if X = Y = C, the closed subsets in X and Y are exactly the
finite subsets, and there are lots of non-polynomial maps from C to C
such that inverse image of a finite subset is finite (describe one!).
Remark 3.5.5 The proof of Proposition 3.3.3 allows us to ‘find’ X from

k[X]. More careful look at the proof however shows that we do not have
a functor from affine algebras to algebraic sets, as ‘recovering’ X from
k[X] is not canonical—it depends on the choice of generators in k[X],
so only ‘recover X up to isomorphism’. The problem here is that our
definition of algebraic sets is not a ‘right one’—it relies on embedding
into some An , and this is something which we want to eventually avoid.
At this stage, we can at least canonically recover X from k[X] as
a topological space. Indeed, we know that as a set, X is in bijection
with the set Specm k[X] of maximal ideals of the algebra k[X]. So if we
want to construct a reasonable quasi-inverse functor G to the functor F,
we could associate Specm k[X] to k[X]. Now make Specm k[X] into a
topological space by considering the topology whose basis consists of all
Xf := {M ∈ Specm | f 6∈ M }. Then x 7→ Mx is a homeomorphism from
X to Specm X. Finally, if α : k[Y ] → k[X] is an algebra homomorphism
define G(α) : Specm k[X] → Specm k[Y ] as follows: if M ∈ Specm k[X]
then G(M ) is the maximal ideal N in k[Y ] containing α−1 (M ). Note
that if we identify X with Specm k[X] as above, and ϕ : X → Y is a
morphism, then ϕ = G(ϕ∗ )—in other words, Mϕ(x) is the maximal ideal
of k[Y ] containing (ϕ∗ )−1 (Mx ).
Example 3.5.6
(i) The notion of a regular function on X and a regular map from
X to k coincide.
(ii) Projection f (T1 , T2 ) = T1 is a regular map of the curve T1 T2 = 1
to k.
(iii) The map f (t) = (t, tk ) is an isomorphism from k to the curve
y = xk .
(iv) The map α(t) = (t2 , t3 ) is a regular map from k to the curve X ⊂
A2 given by x3 = y 2 . This map is clearly one-to-one, but it is not
an isomorphism (even though it is a homeomorphism!) Indeed,
any regular function on X has a representative p(x) + q(x)y in
k[x, y] for some p, q ∈ k[x]. Now α∗ (p(x)+q(x)y) = p(t2 )+q(t2 )t3 ,
which is never equal to t, for example. So α∗ is not surjective.
Moreover, one can see that X is not isomorphic to A1 , since
k[X] 6∼
= k[T ].
Example 3.5.7 Let X be an algebraic set, and G be its finite group of

automorphisms. Then G is also a group of automorphisms of the algebra
A = k[X]. Suppose that char k6 | |G|. Then the invariant algebra AG is
an affine algebra (the only non-trivial thing here is that it is finitely
generated, which can be looked up in [Sh, Appendix].) So there is an
algebraic set Y with k[Y ] = AG , and the regular map π : X → Y with
π ∗ being the embedding of AG into A. This algebraic set Y is called
the quotient of X by G and is denoted X/G. The map π leads to a
natural one-to-one correspondence between the elements of X/G and
the G-orbits on X.
Indeed, we claim that for x1 , x2 ∈ X, one has π(v1 ) = π(v2 ) if and
only if x1 and x2 are in the same G-orbit. Well, if x2 = gx1 , then
f (x1 ) = f (x2 ) for all f ∈ AG = k[Y ], and so π(x1 ) = π(x2 ). Conversely,
if x1 and x2 are not in the same orbit, then let f ∈ k[X] be a function
with f (gT2 ) = 1 and f (gT1 ) = 0 for all g ∈ G (why does it exist?).
1 ∗ G
P
Then the average function S(f ) := |G| g∈G g f belongs to A and
‘separates’ x1 from x2 . So π(x1 ) 6= π(x2 ).
Finally, in view of Remark 3.5.5, the surjectivity of π follows from
the Lying Over Theorem and the Maximality Theorem 2.2.21, if we can
establish that A is integral over AG . Well, for any element f ∈ A, the
coefficients of the polynomial
Y
tN + a1 tN −1 + · · · + aN = (t − g · f ) =: Pf (t)
g∈G
belong to AG , as they are elementary symmetric functions in g · f . On

the other hand Pf (f ) = 0.
3.6 Products
n m
Let X ⊆ A and Y ⊆ A be algebraic sets. Then the cartesian product
X ×Y is an algebraic set in An+m . Indeed, if we identify k[T1 , . . . , Tm+n ]
with k[T1 , . . . , Tn ] ⊗ k[T1 , . . . , Tm ], then it is easy to see that I(X × Y ) =
I(X) ⊗ k[T1 , . . . , Tm ] + k[T1 , . . . , Tn ] ⊗ I(Y ) (check it!).
From Proposition 2.1.2 we get
k[X × Y ] ∼
= k[X] ⊗ k[Y ]. (3.2)
Lemma 3.6.1 Tensor product A ⊗ B of affine k algebras is an affine
k-algebra. Moreover, if A and B are domains, then so is A ⊗ B.
Proof The first statement follows from (3.2) and Proposition 3.3.3. As-
sume A and B are domains and α, α0 ∈ A ⊗ B be such that αα0 = 0.
ai ⊗ bi and α0 =
P 0
ai ⊗ b0i with the sets {bi } and {b0i }
P
Write α =
each linearly independent. Let M be a maximal ideal in A, and ā de-
note a + M ∈ A/M = k. As ( ai ⊗ bi )( a0i ⊗ b0i ) = 0 in A ⊗ B, in
P P
A/M ⊗ B = k ⊗ B = B we have ( āi ⊗ bi )( ā0i ⊗ b0i ) = 0. As B is

P P
domain and the sets {bi } and {b0i } are linearly independent, it follows
that either all ai ∈ M or all a0i ∈ M . Now, recall from Proposition 3.3.3
that A ∼= k[X] for some irreducible variety X. Consider the subvari-
eties Y and Y 0 of X which are zero sets of the functions {ai } and {a0i },
respectively.
Corollary 3.6.2 If X and Y are irreducible then so is X × Y .
Remark 3.6.3 Zariski topology on X × Y is not the product topology

of those on X and Y .
Example 3.6.4 This is a generalization of Example 3.5.6(ii). Let X be

a closed set in An and f ∈ k[X]. Consider the set X 0 ⊆ X × A1 ⊂ An+1
given by the equation Tn+1 f (T1 , . . . , Tn ) = 1. Note that k[X 0 ] ∼ = k[X]f .
Then the projection π(T1 , . . . , Tn , Tn+1 ) = (T1 , . . . , Tn ) defined a regular
map π : X 0 → X. This map defines a homeomorphism between X 0
and the principal open set Xf . This idea will be used to consider a
principal open set as an algebraic variety. In fact, it will turn out that
k[Xf ] = k[X]f .
3.7 Rational functions

In algebraic geometry we need more functions than just globally defined
regular functions on a variety X. In fact, if we were planning to deal
only with affine algebraic sets such globally defined functions would be
‘enough’ in view of Theorem 3.5.1. However, we will see that constant
functions are the only globally defined regular functions on a projective
variety. So, as in complex analysis we are going to allow some ‘poles’
and consider functions which are not defined everywhere on X.
Definition 3.7.1 Let X be an irreducible algebraic set. The field of
fractions of the ring k[X] is denoted k(X) and is called the field of
rational functions on X, its elements being rational functions on X. A
rational function ϕ ∈ k(X) is regular at the point x ∈ X if it can be
written in the form ϕ = fg for f, g ∈ k[X] with g(x) 6= 0. In this case (the
f (x)
well-defined number) g(x) is called the value of ϕ at x and is denoted
ϕ(x).
Note that the set of points on which a rational function ϕ on X is

regular is non-empty and open, and hence dense in X. This set is called
the domain of ϕ. As the intersection of two non-empty open sets in
an irreducible space is non-empty and open again, we can compare a
finite set of rational functions on a non-empty open set. Another useful
remark is that a rational function is uniquely determined by its values
on a non-empty open set. Indeed, if ϕ = 0 on such a set U , then taking
some presentation ϕ = fg for ϕ, we see that f is zero on a non-empty
open set U ∩ (X \ Z(g)), which is dense in X, so f = 0.
Theorem 3.7.2 Rational function ϕ regular at all points of an irre-

ducible affine algebraic set X is a regular function on X.
Proof By assumption, for every x ∈ X we can write ϕ(x) = gfxx (x) (x) for
fx , gx ∈ k[X] with gx (x) 6= 0. Then the zero set in X of the ideal
generated by all functions gx is empty, so by the Nullstellensatz the
ideal equals k[X]. So there exist functions h1 , . . . hn ∈ k[X] and points
Pn
x1 , . . . , xn ∈ X such that i=1 hi gxi = 1. Multiplying both sides of
f
this equality by ϕ (in k(X)) and using the fact that ϕ = gxxi , we get
Pn i
ϕ = i=1 hi fxi , so ϕ ∈ k[X].
The subring of K(X) consisting of all functions regular at the point

x ∈ X is denoted Ox and called the local ring of x. Note that Ox ∼ =
k[X]Mx , the localization of k[X] at the maximal ideal Mx . So Ox is a
local ring in the sense of commutative algebra with the maximal ideal
mx consisting of all rational functions representable in the form fg with
f (x) = 0 6= g(x). Now Theorem 3.7.2 can be interpreted as
k[X] = ∩x∈X Ox . (3.3)
Informally speaking the local ring Ox describes what happens ‘near the
point x’. This becomes a little more clear if we note that Ox is the
same as the stalk of rational functions at x: the elements of the stalk are
germs of rational functions at x. One can think of germs as equivalence
classes of pairs (U, f ), where U is an open set containing x, f is a rational
function regular at all points of U , and (U, f ) ∼ (V, g) if there is an open
set W ⊂ U ∩ V and f |W = g|W .
Now, let X ⊆ An be an arbitrary (not necessarily irreducible) alge-
braic set and U ⊆ X be an open subset. A function f : U → k is regular
if for each x ∈ U there exist g, h ∈ k[T1 , . . . , Tn ] such that h(x) 6= 0
and f = hg in some open neighborhood of x. The algebra of all regular
functions on U is denoted OX (U ). Now Ox is defined as the stalk of
functions regular in neighborhoods of x.
Now, let X ⊆ An be an affine algebraic set and 0 6= f ∈ k[X]. Then
the elements of the localization k[X]f can be considered as regular func-
tions on the principal open set Xf (we do imply here that different
elements of k[X]f give different functions—check!) We claim that these
are precisely all regular functions on Xf :
Theorem 3.7.3 k[X]f is the algebra of regular functions on Xf .
Proof Let g be a regular function on Xf . So we can find an open

covering of Xf such that on each element U of this covering g equals
a
b for a, b ∈ k[T1 , . . . , Tn ] (with b(x) 6= 0 for all x ∈ U ). But principal
open sets form a basis of Zariski topology on An , and the topology is
noetherian. So we may assume that Xf = Xg1 ∪ · · · ∪ Xgl and g = abii
on Xgi for i = 1, . . . , l. Then Xgi ⊆ Xbi . From now on we consider all
functions as functions on X via restriction. By the Nullstellensatz, for
each i, we have gini = bi hi for some ni ∈ Z≥0 and hi ∈ k[X]. Note that
hi (x) 6= 0 for any x ∈ Xgi , so
ai ai hi ai hi
= = ni .
bi bi hi gi
on Xgi . As Xgi = Xgni , renaming ai hi as ai and gini as gi we have that
i
g = agii on Xgi .
a
Now, on Xgi ∩ Xgj = Xgi gj we have agii = gjj , whence ai gj − aj gi = 0,
therefore (ai gj − aj gi )gi gj = 0 everywhere on X. So ai gi gj2 = aj gi2 gj .
Moreover, on Xgi we have agii = agi 2gi . Renaming ai gi as ai and gi2 as
i
gi , we are reduced to the case g = agii on Xgi and ai gj = aj gi on X.
Now the condition Xf = Xg1 ∪ · · · ∪ Xgl and the Nulstellensatz imply
f n = i ci gi for some ci ∈ k[X] for some n. So
P
aj aj X X aj gi ci X ai gj ci X
gf n |Xgj = f n = ci gi = = = ai ci .
gj gj i i
gj i
gj i
Xgi ’s cover Xf , it follows that gf n =
P
Since
P i ci ai on Xf . So g =
i ci ai
fn ∈ k[X]f , as required.
3.8 Projective n-space

The objects that algebraic geometry can study are much more diverse
than just affine algebraic set. To extend our horizons we now demon-
strate how projective algebraic sets can be studied. Algebraically, this
just means considering homogeneous polynomials instead of all polyno-
mials.
Define the projective n-space Pn as the set of equivalence classes on
n+1
k \ {(0, . . . , 0)} with respect to the following equivalence relation:
(x0 , x1 , . . . , xn ) ∼ (y0 , y1 , . . . , yn ) if and only if there exists an element
c ∈ k × such that yi = cxi for all i = 0, 1, . . . , n.
Thus every point of Pn has n + 1 coordinates x0 , . . . , xn , which are
only defined up to a non-zero scalar multiple. To emphasize this fact we
will refer to the coordinates of this point as the homogeneous coordinates
and denote them by
(x0 : x1 : · · · : xn ).
If we want to consider subsets of Pn which are zero sets of polynomi-
als in the homogeneous coordinate functions S0 , S1 , . . . , Sn we have to
require that these polynomials are homogeneous.
Definition 3.8.1 Let S be a set of homogeneous polynomials in k[S0 , S1 , . . . , Sn ].

A zero of the set S is an element (x0 : x1 : · · · : xn ) of Pn such that
f (x0 , x1 , . . . , xn ) = 0 for all f ∈ S. The zero set of S is the set Z(S) of
all zeros of S. An algebraic set in Pn (or projective algebraic set) is the
zero set of some set of homogeneous polynomials S ⊆ k[S0 , S1 , . . . , Sn ],
in which case S is called a set of equations of the algebraic set.
Note that Z(S) = Z((S)), where (S) is the ideal of k[S0 , S1 , . . . , Sn ]
generated by S. Therefore every algebraic set is the zero set of some
homogeneous ideal. Now, by Hilbert’s Basis Theorem, every algebraic
set is the sero set of a finite set of homogeneous polynomials.
As in the affine case, one proves that the algebraic sets are closed sets
of a topology on Pn , which again is called the Zariski topology. Principal
open sets form a base of this topology.
The map
I : {algebraic sets in Pn } → {homogeneous ideals in k[S0 , . . . , Sn ]}
3.8 Projective n-space 33
is defined in the obvious way (you need to check that I(X) is homoge-
neous!).
Definition 3.8.2 The ideal M0 of k[S0 , . . . , Sn ] generated by S0 , . . . , Sn

is called the superfluous ideal.
The following projective version of the Nullsellensatz follows easily
from the classical one.
Theorem 3.8.3 (Projective Nullstellensatz) The maps I and Z

induce an order-reversing bijection between algebraic sets in Pn and
non-superfluous homogeneous radical ideals in k[S0 , . . . , Sn ]. Under this
correspondence, irreducible algebraic sets correspond to the prime ideals.
Let Ui ⊂ Pn be the subset consisting of all points with non-zero ith
homogeneous coordinate. This is the principal open set corresponding
to the function Si . We call the Ui (the ith) affine open set in Pn . The
terminology is justified by the following. The map
αi : (x0 , . . . , xn ) 7→ (x0 /xi , . . . , xi−1 /xi , xi+1 /xi , . . . , xn /xi )
is a bijection between Ui and An . We will refer to the functions
Tj : (x0 , . . . , xn ) 7→ xj /xi , (j = 0, . . . , i − 1, i + 1, . . . , n)
as the affine coordinates on Ui .
We claim that αi is not just a bijection but a homeomorphism between
Ui and An . Indeed, to each polynomial f (T0 , . . . , Ti−1 , Ti+1 , . . . , Tn ) we
associate its homogenization
fˆ(S0 , . . . , Sn ) := Sideg f f (S0 /Si , . . . , Si−1 /Si , Si+1 /Si , . . . , Sn /Si ),
which is clearly a homogeneous polynomial in S0 , . . . , Sn . Now, if X in
An is the zero set of polynomials f1 , . . . , fm ∈ k[T0 , . . . , Ti−1 , Ti+1 , . . . , Tn ],
then
α−1 (X) = Ui ∩ Z(fˆ1 , . . . , fˆm ).
We note in passing, that Z(fˆ1 , . . . , fˆm ) is the closure in Pn of α−1 (X)
(why?). Conversely, to each homogeneous polynomial g(S0 , . . . , Sn ) we
associate the polynomial
ḡ(T0 , . . . , Ti−1 , Ti+1 , . . . , Tn ) := g(T0 , . . . , Ti−1 , 1, Ti+1 , . . . , Tn ).
Now
α(Z(g1 , . . . , gl ) ∩ Ui ) = Z(ḡ1 , . . . , ḡl ).
Lemma 3.8.4 (Affine Criterion) Let X be a topological space with
an open cover X = ∪i∈I Ui , and Y ⊆ X. Then Y is closed if and only
Y ∩ Ui is closed in Ui for all i. In particular, a subset Y of Pn is closed
if and only if its intersection Y ∩ Ui with the ith affine open set is closed
in Ui for all i.
Proof The ‘only-if’ part is obvious. For the ‘if’ part, by assumption
each Y ∩ Ui = Zi ∩ Ui for some closed set Zi in Pn . It suffices to check
that
Y = ∩i∈I (Zi ∪ (Pn \ Ui )).
Well, let y ∈ Y and i ∈ I. Either y ∈ Ui and then y ∈ Y ∩ Ui ⊂ Zi , or
y ∈ Pn \ Ui . Conversely, if y ∈ Zi ∪ (Pn \ Ui ) for all i. As Pn = ∪Ui , there
is an i with y ∈ Ui . Then y 6∈ Pn \Ui , hence y ∈ Zi , and x ∈ Zi ∩Ui ⊂ Y .
3.9 Functions
p(S0 ,...,Sn )
A rational expression f = can be considered as a function on
q(S0 ,...,Sn )
n
P (defined at the points where q(S0 , . . . , Sn ) 6= 0) only if p and q are
homogeneous of the same degree, in which case we will refer to f as a
rational function of degree 0. Let X ⊂ Pn be a projective algebraic set,
x = (x0 , . . . , xn ) ∈ X, and f = pq be of degree 0. If q(x0 , . . . , xn ) 6= 0,
then we say that f is regular at x. If a degree 0 rational function is
regular at x, then it is also regular on some neighborhood of x. For any
set Y ⊆ Pn , a function f on Y is called regular if for any x ∈ Y there
exists a rational function g regular at x and such that f = g on some
open neighborhood of x in Y . If U is an open subset of X we write
OX (U ) for the set of all regular functions on U .
We will prove later that the only functions regular on projective alge-
braic sets are constants. This underscores the importance of considering
rational functions regular only on some open subsets.
Let U be an open subset of Pn contained in some affine open set
Ui . Then U is also open in Ui , which is canonically identified with An .
We claim that OPn (U ) = OAn (U ). Indeed, assume for example that
i = 0, and let f ∈ OPn (U ). This means that there is an open cover U =
p (S ,...,S )
W1 ∪· · ·∪Wl in Pn and rational functions qjj (S00,...,Snn) defined on Wj such
p p (1,T ,...,T )
that f = qjj on Wj , j = 1, . . . , l. Then we also have f = qjj (1,T11,...,Tnn) on
Wj , where T1 , . . . , Tn are the affine coordinates on U0 . Conversely, let
f ∈ OAn (U ). This means that there is an open cover U = V1 ∪ · · · ∪ Vm
3.10 Product of projective algebraic sets 35
gj (T1 ,...,Tn ) gj
of U and rational functions hj (T1 ,...,Tn ) defined on Vj such that f = hj
deg hj
S0 ĝj
on Vj , j = 1, . . . , m. Now, on Vj we can also write f = deg gj , where
S0 ĥj
ĝj and ĥj are homogenizations.
Let X ⊆ Pn be a projective algebraic set, and U0 , . . . , Un be the affine
open sets in Pn . Put Vi := X ∩Ui . Then X = V0 ∪· · ·∪Vl is an open cover
of X. Moreover, Vi an affine algebraic set in Ui , and Ui is canonically
identified with An . Let U be an open subset of X which is contained
in some Vi . Then U is an open subset of Vi . The argument as in the
previous paragraph can be modified to prove the following more general
result: a function on U is regular in the sense of the projective algebraic
set X if and only if it is regular in the sense of the affine algebraic set
Vi , i.e. OX (U ) = OVi (U ).
3.10 Product of projective algebraic sets

Let X ⊂ P and Y ⊂ Pm be projective algebraic sets. We would like
n
to consider X × Y as a projective algebraic set in a natural way. For

example, we could have X = Pn and Y = Pm . It is quite clear that there
is no natural identification of Pn × Pm with Pn+m (play with that!). But
there is a natural Segre embedding of Pn × Pm into P(n+1)(m+1)−1 :
ϕ : Pn × Pm → P(n+1)(m+1)−1 ,
((T0 , . . . , Tn ), (S0 , . . . , Sm )) 7→ (T0 S0 , . . . , T0 Sm , . . . , Tn S0 , . . . , Tn Sm )
It is easy to see that ϕ is injective. We next show that im ϕ is closed
in P(n+1)(m+1)−1 . Let wij , 0 ≤ i ≤ n, 0 ≤ j ≤ m be the homogeneous
coordinates in P(n+1)(m+1)−1 . We claim that im ϕ is the zero set of the
following equations:
wij wkl = wkj wil (0 ≤ i, k ≤ n, 0 ≤ j, l ≤ m). (3.4)
That all points of im ϕ satisfy these equations is clear. Conversely, if the
numbers wij satisfy these equations, and wkl 6= 0, then
(· · · : wij : . . . ) = ϕ(x, y)
where x = (w0l : · · · : wnl ) and y = (wk0 : · · · : wkm ).
So, we have proved that the image of Pn × Pm under the Segre embed-
ding is a projective algebraic set, and this is what we will understand by
the product of Pn and Pm . More generally, let X be an algebraic set in
Pn and Y be an algebraic set in Pm . By the product of X and Y we un-
derstand ϕ(X × Y ), which we show to be algebraic. Well, if X is given
by the equations Fα (T0 , . . . , Tn ) = 0 and Y is given by the equations
Gβ (S0 , . . . , Sm ) = 0, then X × Y is the zero set of the equations (3.5)
together with Fα (w0j , . . . , wnj ) for 1 ≤ j ≤ m and Gβ (wi0 , . . . , wim ) for
1 ≤ i ≤ n.
3.11 Example: Grassmann varieties and flag varieties

Let V be an n-dimensional vector space. As a set, the Grassmann variety
Gr (V ) (or Gr (n)) is just the set of all r-dimensional (linear) subspaces
in V . However, we need to explain how is Gr (V ) a projective algebraic
set. Of course, we already know that for r = 1 when Gr (V ) is nothing
but the projective space P(V ) = Pn−1 . In general we are going to realize
Gr (V ) as an algebraic set in the projective space P(Λr (V )).
Define the map
ψ : Gr (V ) → P(Λr (V ))
as follows. Let l1 , . . . , lr be a basis of a subspace L ⊂ V . Then ψ(L) is

defined to be the span of the vector l1 ∧ · · · ∧ lr ∈ Λr (V ). It is easy to
check that ψ is a well defined embedding. We claim that the image of
ψ is an algebraic set. In order to see that, let us fix a basis {v1 , . . . , vn }
of V . Then the basis of Λr (V ) is
{vi1 ∧ · · · ∧ vir | 1 ≤ i1 < · · · < ir ≤ n}.
Denote the vi1 ∧ · · · ∧ vir -coefficient of l1 ∧ · · · ∧ lr by µi1 ...ir . Then

the homogeneous coordinates of ψ(L) are (· · · : µi1 ...ir : . . . ). These
homogeneous coordinates are called the Plükker coordinates of L. We
accept the following convention: given a collection of numbers {µi1 ...ir |
1 ≤ i1 < · · · < ir ≤ n} we assume that µi1 ...ir are also defined for any
i1 , . . . , ir with 1 ≤ i1 , . . . , ir ≤ n in such a way that after two indices are
interchanged, µi1 ...ir gets multiplied by −1; in particular, if two indices
are the same, it is zero.
With these assumptions the Plükker coordinates can be described as
Pn
follows. Write li = j=1 aij vj . Then µi1 ...ir is the determinant of the
matrix formed by the columns of A := (aij ) with indices i1 , . . . , ir .
Theorem 3.11.1 Numbers µi1 ...ir are Plükker coordinates of some r-

dimensional subspace L ⊂ V if and only if they are not simultaneously
zero and if for all i1 , . . . , ir+1 , j1 , . . . , jr−1 the following relation (called
3.11 Example: Grassmann varieties and flag varieties 37
Plükker relation) holds:
r+1
X
(−1)k µi1 ...ibk ...ir+1 µik j1 ...jr−1 = 0.
k=1
Proof Expanding the determinant µik j1 ...jr−1 along the first column, we
obtain
r
X
µik j1 ...jr−1 = asik Ns ,
s=1
where Ns does not depend on k. Thus, it suffices to prove that

r+1
X
(−1)k µi1 ...ibk ...ir+1 asik = 0 (3.5)
k=1
for all s. Add the sth row to A to obtain an (r + 1) × n matrix As .

Then the left hand side of (3.5) is, up to a sign, the expansion of the
determinant of the matrix formed by the columns of As with indices
i1 , . . . , ir+1 along the last row. But this determinant is zero.
Conversely, assume that µi1 ...ir are not simultaneously zero and the
Plükker relations hold. It suffices to prove that there exists an r × n
matrix A such that
µi1 ...ir = Mi1 ...ir (1 ≤ i1 , . . . , ir ≤ n), (3.6)
where Mi1 ...ir is the minor formed by the columns of A with indices
i1 , . . . , ir . We may assume that µ1...r = 1. We will look for A in the
form
 
1 0 . . . 0 a1,r+1 . . . a1n
0 1 . . . 0 a2,r+1 . . . a2n 
..  .
 
. . .. .. .. ..
 .. .. . . . . . 
0 0 ... 1 ar,r+1 ... arn
Note that for j > r we have M1...î...rj = (−1)r−i aij . Thus , we must
set aij = (−1)r−i µ1...î...rj , in which case the equality (3.6) holds at least
for the sets {i1 , . . . , ir } which differ from {1, . . . , r} in no more than one
element.
Now it remains to prove that (3.6) holds if the set {i1 , . . . , ir } differs
from {1, . . . , r} in m elements for any m. We use induction on m. We
may assume that i1 6∈ {1, . . . , r}. Then, using the Plükker relations, we
get
r
X
µi1 ...ir = µ1...r µi1 ...ir = (−1)k+1 µi1 1...k̂...r µki2 ...ir . (3.7)
k=1
On the other hand, it follows from the first part of the theorem that the
same condition holds for the minors of A:
r
X
Mi1 ...ir = (−1)k+1 Mi1 1...k̂...r Mki2 ...ir . (3.8)
k=1
By the induction hypothesis, the right hand sides of (3.7) and (3.8)
coincide. Therefore Mi1 ...ir = µi1 ...ir .
A flag in the n-dimensional vector space V is a chain
0 ⊂ V1 ⊂ V2 ⊂ · · · ⊂ Vn = V
of subspaces with dim Vi = i for all i = 1, . . . , n. Let F(V ) be the

set of all flags in V . This set can be given a natural structure of a
projective algebraic set called flag variety. Note that Vi ∈ Gi (V ), so we
can consider F(V ) as a subset of G1 (V ) × · · · × Gn (V ), and we claim
that this is a closed subset.
Indeed, it suffices to prove that the condition for Vd to be contained
in Vd+1 is a closed condition for each d. In checking that we may forget
about other spaces and work in P(Λd (V )) × P(Λd+1 (V )). Let us apply
Affine Criterion. The open covering we are going to use is the direct
products of the affine open sets in P(Λd (V )) and P(Λd+1 (V )). The
affine open sets in P(Λd (V )) are given by conditions µi1 ...id 6= 0. As
they are all the same we may work with the set U given by µ1...r 6= 0.
Then Vd ∈ U if and only if Vd is spanned by the vectors of the form
vi + j=d+1 aij vj , i = 1, . . . , d. In fact, U ∩ Gd (V ) ∼
Pn
= Ad(n−d) and the
aij can be considered as the affine coordinates on U ∩ Gd (V ). Now,
let U 0 be the affine open set in P(Λd+1 (V )) containing Vd+1 given by
µi1 ...id+1 6= 0. As Vd ⊂ Vd+1 , we must have that i1 = 1, . . . , id = d, for
otherwise the intersection with U × U 0 is empty. We may also assume
without loss of generality that id+1 = d + 1. Now, Vd+1 is spanned by
Pn
the vectors of the form vi + j=d+2 bij vj , i = 1, . . . , d+1. In fact, the bij
can be considered as the affine coordinates on U 0 ∩ Gd+1 (V ). Now the
condition that Vd is contained in Vd+1 can be written by the polynomial
equations aij = bij + ai,d+1 bd+1,i for all 1 ≤ i ≤ d and d + 2 ≤ j ≤ n.
3.12 Example: Veronese variety 39
3.12 Example: Veronese variety
Consider all homogeneous polynomials of degree m in S0 , S1 , . . . , Sn .
n+m

They form a vector space of dimension m . The corresponding pro-
jective space is Pνn,m where νn,m := n+m m − 1. To each point of Pνn,m
there corresponds a hypersurface of degree m in Pn (since proportional
polynomials define the same hypersurface).
Denote the homogeneous coordinates in Pνn,m by vi0 ...in for all tuples
(i0 , . . . , in ) of non-negative integers with i0 + · · · + in = m. Consider the
map αm : Pn → Pνn,m , defined by
vi0 ...in (αm ((a0 : · · · : an ))) = ai00 . . . ainn . (3.9)
The map is well-defined, as among the monomials in the right hand side
of (3.9) there are am
i which all turn into 0 only if all ai = 0. The map
αm is clearly injective. It is called Veronese map, and αm (Pn ) is called
Veronese variety.
Formulas (3.9) imply that all points of the Veronese variety satisfy
equations
vi0 ...in vj0 ...jn = vk0 ...kn vl0 ...ln

(3.10)
if i0 + j0 = k0 + l0 , . . . , in + jn = kn + ln .
Conversely, it follows from the relations (3.10) that at least one of the
coordinates of the form v0...m...0 is non-zero. Indeed, assume otherwise,
and prove by induction on the amount k of non-zeros among {i0 , . . . , in }
that all vi0 ...in = 0. The induction base k = 1 follows from our assump-
tion. On the other hand, let k ≥ 2 and assume that the statement is
true for k − 1. Let ir be the minimal non-zero element in {i0 , . . . , in }
and is be the minimal non-zero element of {i0 , . . . , in } \ {ir }. Now, the
relation
vi20 ...ir ...is ...in = vi0 ...0...is +ir ...in vi0 ...ir ...is −ir ...in
is among the relations (3.9). By the inductive assumption, the right

hand side of it is zero, so vi0 ...in is also zero, completing the induction
step.
Now, let, for example, vm0...0 6= 0. Then our point with homogeneous
coordinates (vi0 ...in ) is the image under the Veronese map of the point
with coordinates
u0 = vm0...0 , u1 = vm−1,1,0...0 , . . . , un = vm−1,0...0,1 .

Indeed, it suffices to check that
(vm0...0 )i0 (vm−1,1,0...0 )i1 . . . (vm−1,0...0,1 )in
m−1 = vi0 ...in .
vm0...0
or, equivalently,
(vm0...0 )i0 −m+1 (vm−1,1,0...0 )i1 . . . (vm−1,0...0,1 )in = vi0 ...in . (3.11)
We prove this by induction on the lexicographical order on the tuples

(i0 . . . in ). For the highest tuple (m0 . . . 0) the result is obvious. Every
other (i0 . . . in ) has some ir 6= 0. Now,
vi0 ...ir ...in vm0...0 = vi0 +1...ir −1...in vm−1...1...0 . (3.12)
If vm−1...1...0 = 0, it follows that vi0 ...in = 0, in which case (3.11) is clear.

Otherwise, substituting (3.12) into (3.11), we reduce (3.11) for (i0 , . . . in )
to (3.11) for (i0 + 1, . . . , ir − 1, . . . , in ), which is true by induction.
ai0 ...in ui00 . . . uinn be a form of degree m and H be a hy-
P
Let F =
persurface in Pn defined by the equation F = 0. Then αm (H) is the
intersection of αm (Pn ) with the hyperplane given by the equation
X
ai0 ...in vi0 ...in = 0.
Let us now concentrate on the special case
α3 : P1 → P3 : (a0 : a1 ) 7→ (a30 : a20 a1 : a0 a21 : a31 ).
The corresponding Veronese variety C is called the twisted cubic. It is

described by the equations
F0 = F1 = F2 = 0, (3.13)
where
2 2
F0 = v30 v12 − v21 , F1 = v21 v12 − v30 v03 , F2 = v21 v03 − v12 .
The twisted cubic consists of all points of the form (1 : c : c2 : c3 ) for

c ∈ k together with the point (0 : 0 : 0 : 1). Let Qi be the hypersurfaces
described by Fi = 0. Then C = Q0 ∩ Q1 ∩ Q2 , but C 6= Qi ∩ Qj for any
two hypersurfaces Qi and Qj . In fact the following beautiful geometric
fact is true: the intersection Qi ∩ Qj equals C ∪ Lij for some (projective)
line Lij (it is easy to see that no line is contained in C).
In order to prove this we consider a more general problem. For λ =
(λ0 : λ1 : λ2 ) ∈ P2 define the hypersurface Qλ by Fλ , where
Fλ := λ0 F0 + λ1 F1 + λ2 F2 .
3.13 Problems 41
We claim that for λ 6= µ, one has Qλ ∩ Qµ = C ∪ Lλ,µ for some line Lλ,µ .
Note that the equations (3.13) are equivalent to the requirement that
the matrix

v30 v21 v12
v21 v12 v03
has rank less than 2. Now note that Fλ is the determinant of the matrix
 
v30 v21 v12
v21 v12 v03  .
λ 2 λ 1 λ0
So the locus outside of C of Fλ = Fµ = 0 is the rank ≤ 2 locus of the
matrix
 
v30 v21 v12
v21 v12 v03 
 ,
 λ2 λ1 λ0 
µ2 µ1 µ0
which as λ and µ are linearly independent is the same as the locus of

v30 v21 v12 v21 v12 v03

λ2
λ1 λ0 = λ2 λ1 λ0 = 0,
µ µ1 µ0 µ2 µ1 µ0
2
which is a line.
3.13 Problems
Problem 3.13.1 True or false? Let I, J be ideals in k[T1 , . . . , Tn ]. Then

Z(I) ∪ Z(J) = Z(IJ).
√
Solution.
√ True. By the Nullstellensatz,
√ it suffices
√ to prove that I ∩J =
√IJ. Well, IJ n⊂ I ∩ J implies IJ ⊂ I ∩ J. Conversely, let x ∈
I ∩ J. Then x ∈ I ∩ J, whence x2n ∈ IJ.
Problem
√ 3.13.2 True or false? Let I, J be ideals in k[T1 , . . . , Tn ]. Then
√
I ∩ J = IJ.
Solution. True. See the previous problem.
Problem 3.13.3 Let I and J be ideals of A = C[x, y] and Z(I)∩Z(J) =

∅. Show that A/(I ∩ J) ∼
= A/I × A/J.
Solution. In view of the Chinese Remainder Theorem, we need only to
show that I + J = A. Otherwise, let M be a maximal ideal containing
I + J. By the Nullstellensatz, M = Ma for some a ∈ C2 . Then a ∈
Z(I) ∩ Z(J).
Problem 3.13.4 True or false? Any decreasing sequence of algebraic

sets in An stabilizes.
Solution. True by the Nullstellensatz and Hilbert Basis Theorem.
Problem 3.13.5 True or false? Any increasing sequence of algebraic

sets in An stabilizes.
Solution. False. Take ”increasing sets of points”.
Problem 3.13.6 If X = ∪Uα is an open covering of an algebraic set,

then X = Uα1 ∪ · · · ∪ Uαl for some α1 , . . . , αl .
Solution. Otherwise we would have an infinite strictly decreasing se-

quence of closed subsets, which contradicts Problem 3.13.4.
Problem 3.13.7 True or False?

(i) {(x, y) ∈ A2 | x2 + y 2 = 1} is homeomorphic to k (in Zariski
topology).
(ii) The set k \ {(0)} with induced Zariski topology is not homeomor-
phic to any variety.
Solution. (i) True. Our variety has the same cardinality as k and cofinite
topology, see Lemma 2.1.1 (even characteristic 2 is O.K., because then
Z(x2 + y 2 − 1) = Z(x + y − 1)).
(ii) False. This set and k have the same cardinality and cofinite topol-
ogy.
Problem 3.13.8 True or false? A system of polynomial equations
f1 (T1 , . . . , Tn ) = 0
..
.
fm (T1 , . . . , Tn ) = 0
over k has no solutions in An if and only if 1 can be expressed as a linear

P
combination 1 = i pi fi with polynomial coefficients pi .
3.13 Problems 43
Solution. True. The first condition is equivalent to (f1 , . . . , fm ) = k[T ],
in view of the Nullstellensatz.
Problem 3.13.9 Let char k 6= 2. Decompose

Z(x2 + y 2 + z 2 , x2 − y 2 − z 2 + 1)
into irreducible components.
Solution. An easy calculation shows that Z(x2 +y 2 +z 2 , x2 −y 2 −z 2 +1)
equals
√ √
Z(x = i/ 2, y 2 + z 2 = 1/2) ∪ Z(x = −i/ 2, y 2 + z 2 = 1/2),
union of two irreducible sets, since y 2 + z 2 = 1/2 is an irreducible poly-
nomial.
Problem 3.13.10 True or false? The Zariski topology on Am+n is the

product topology of the Zariski topologies on Am and An .
Solution. False. Consider the case m = n = 1.
Problem 3.13.11 Let k have characteristic p > 0, and Fr : k → k, a 7→

ap be the Frobenius homomorphism. True or false:
(i) Fr is a homeomorphism in the Zariski topology.
(ii) Fr is an isomorphism of algebraic sets.
Solution. (i) is true, as Fr is a bijection. (ii) is false as Fr∗ is not an
isomorphism.
Problem 3.13.12 Prove that the hyperbola xy = 1 and k are not

isomorphic.
Solution. If ψ : k[x, y]/(xy − 1) → k[T ] is an isomorphism, then ψ(x)
and ψ(y) must be invertible, which leads to a contradiction.
Problem 3.13.13 For the regular map f : A2 → A2 , (x, y) 7→ (x, xy)

describe im f . Is the image dense in A2 ? Open? Closed?
Solution. The image is A2 \ {(0, b) | b 6= 0}. It is dense because it
contains a non-empty open set x 6= 0. So it is not closed. It is also not
open, as the origin belongs to the closure of the complement C (in fact,
I(C) = (x)).
Problem 3.13.14 Let X consist of two points. Prove that k[X] ∼

= k ⊕k.
Solution. Use the Nullstellensatz and the Chinese Remainder Theorem
(cf. Problem 3.13.3).
Problem 3.13.15 Describe all automorphisms of the algebraic set k.

Solution. All automorphisms are linear of the form x 7→ ax + b with
a 6= 0. This follows by considering automorphisms of k[T ]. By the way,
the automorphism group is isomorphic to the semidirect product of k ×
and k.
Problem 3.13.16 The graph of a morphism ϕ : X → Y of affine

algebraic sets is a closed set in X × Y isomorphic to X.
Solution. Let s1 , . . . , sn be coordinate functions on Y . Then the graph
is the zero locus of the functions ϕ∗ (si ) ⊗ 1 − 1 ⊗ si ∈ k[X] ⊗ k[Y ] =
k[X × Y ]. Next, check that the maps x 7→ (x, f (x)) and (x, f (x)) 7→ x
are morphisms between X and the graph which are inverse to each other.
Problem 3.13.17 Let ϕ : X → Y be a morphism of affine algebraic

sets. Show that inverse image of a principal open set in Y is a principal
open set in X.
Solution. ϕ−1 (Yf ) = Xϕ∗ (f ) .
Problem 3.13.18 Let X, X 0 be topological spaces.

(i) A subspace Y ⊆ X is irreducible if and only if Ȳ is irreducible.
(ii) If ϕ : X → X 0 is a continuous map and X is irreducible, then
ϕ(X) is irreducible.
Solution. See Humphreys.
Problem 3.13.19 Let ϕ : X → Y be a regular map. Then ϕ(X) is

dense in Y if and only if ϕ∗ is injective. Give an example when ϕ(X) is
dense in Y but ϕ(X) 6= Y .
Solution. I(im ϕ) = {g ∈ k[Y ] | g(ϕ(x)) = 0 for any x ∈ X} = {g ∈
k[Y ] | ϕ∗ (g) = 0} = ker ϕ∗ . Now the result follows from Z(I(im ϕ)) =
im ϕ. For the example see Problem 3.13.13.
Problem 3.13.20 Let X, Y ⊂ Ar be closed subsets, and ∆ ⊂ A2r be

the diagonal, i.e. a subset given by equations T1 = S1 , . . . , Tr = Sr . If
z ∈ X ∩ Y define ϕ(z) = (z, z). Prove that ϕ defines an isomorphism
from X ∩ Y onto (X × Y ) ∩ ∆.
3.13 Problems 45
Solution. (x, y) 7→ x defines the inverse morphism.
Problem 3.13.21 True or false? Let X be an affine algebraic set with

irreducible components X1 , . . . , Xl . Then a function f on X is in k[X]
if and only if f |Xi ∈ k[Xi ] for all i.
Solution. This is actually false! Let X = X1 ∪ X2 , where X1 is the line
in A2 given by x = 0, and X2 ⊂ A2 is the parabola x = y 2 . Consider
the function f which is 0 on X1 , and which maps the point (y 2 , y) of
X2 to y. Then clearly f |X1 and f |X2 are regular. Now assume that
there is a polynomial F (x, y) with F |X = f . Since F |X1 =0, it follows
that F (x, y) = xg1 (x, y) + x2 g2 (x, y) + . . . . Now, F |X2 = f |X2 gives
y = F (y 2 , y) = y 2 g1 (y 2 , y) + y 4 g2 (y 2 , y) + . . . , which is impossible by
degrees.
4
Varieties
4.1 Affine varieties

In this section we will define affine varieties which can be thought of as
a ‘coordinate-free version’ of affine algebraic sets and functions on them.
Definition 4.1.1 A sheaf of functions on a topological space X is a

function F which assigns to every non-empty open subset U ⊂ X a
k-algebra F(U ) of k-valued functions on U (with respect to the usual
point-wise operations) such that the following two conditions hold:
(i) If U ⊂ V are two non-empty open sets and f ∈ F(V ), then the
restriction f |U ∈ F(U ).
(ii) Given a family of open sets Ui , i ∈ I, covering U and functions
fi ∈ F(Ui ) for each i ∈ I, such that fi and fj agree on Ui ∩ Uj ,
there must exist a function f ∈ F(U ) whose restriction to Ui
equals fi .
Definition 4.1.2 A topological space X together with a sheaf of func-

tions OX is called a geometric space. We refer to OX as the structure
sheaf of the geometric space.
Definition 4.1.3 Let (X, OX ) and (Y, OY ) be geometric spaces. A

morphism
f : (X, OX ) → (Y, OY )
is a continuous map f : X → Y such that for every open subset U of Y
and every ϕ ∈ OY (U ) the function
f ∗ (ϕ) := ϕ ◦ f
belongs to OX (f −1 (U )).
46
4.1 Affine varieties 47
Remark 4.1.4 We will often use a shorthand f : X → Y for the
morphism f : (X, OX ) → (Y, OY ).
Example 4.1.5 Let X be an affine or a projective algebraic set. To

each non-empty open subset U ⊂ X we assign the ring OX (U ) which
consists of all regular functions on U . Then (X, OX ) is a geometric
space. Moreover the notion of a morphism agrees with the one we had
before (think about it!).
Let (X, OX ) be a geometric space and Z be a subset of X with induced

topology. We can make Z into a geometric space by defining OZ (V ) for
an open V ⊂ Z as follows: f : V → k is in OZ (V ) if and only if there
exists an open covering V = ∪i Vi in Z such that for each i we have
f |Vi = gi |Vi for some gi ∈ OX (Ui ) where Ui is an open subset of X
containing Vi . It is not difficult to see that OZ is a sheaf of functions on
Z (see it!). We will refer to it as the induced structure sheaf and denote
it by OX |Z. Note that if Z is open in X then a subset V ⊂ Z is open
in Z if and only if it is open in X, and OX (V ) = OZ (V ).
Let X be a topological space and X = ∪i Ui be its open cover. Given
sheaves of functions OUi on Ui for each i, which agree on each Ui ∩ Uj ,
we can define a natural sheaf of functions OX on X by ‘gluing’ the OUi .
Let U be an open subset in X. Then OX (U ) consists of all functions on
U , whose restriction to each U ∩ Ui belongs to OUi (U ∩ Ui ).
If x ∈ X we denote by evx the map from functions on X to k obtained
by evaluation at x:
evx (f ) = f (x).
Definition 4.1.6 A geometric space (X, OX ) is called an affine (alge-

braic) variety if the following three conditions hold:
(i) k[X] := OX (X) is a finitely generated k-algebra, and the map
X → Homk−alg (k[X], k), x 7→ evx
is a bijection.
(ii) For each 0 6= f ∈ k[X] the set
Xf := {x ∈ X | f (x) 6= 0}
is open, and every non-empty open set in X is a union of some

Xf .
(iii) OX (Xf ) = k[X]f .
48 Varieties
Example 4.1.7 It follows from the results of chapter 3 (in particular,
Theorem 3.7.3) that affine algebraic sets with sheaves of regular func-
tions are affine varieties. We claim that, conversely, every affine variety
is isomorphic (as a geometric space) to an affine algebraic set with the
sheaf of regular functions. Indeed, let (X, OX ) be an affine variety. Since
k[X] is a finitely generated algebra of functions, we can write
k[X] = k[T1 , . . . , Tn ]/I
for some radical ideal I. By the property (i) of affine varieties and
the Nulstellensatz, we can identify X with Z(I) as a set, and k[X]
with the regular functions on Z(I). The Zariski topology on Z(I) has
the principal open sets as its base, so it now follows from (ii) that the
identification of X and Z(I) is a homeomorphism. Finally, by (iii),
OX (Xf ) and the regular functions on the principal open set Xf are also
identified. This is enough to identify OX (U ) with regular functions on
U for any open set U , as regularity is a local condition.
Remark 4.1.8 The argument of Example 4.1.7 shows that the affine
variety can be recovered completely from its algebra A := k[X] of reg-
ular functions, and conversely. We make it precise as follows. Define a
functor F from the category of affine varieties to the category of affine
algebras via F(X) = k[X] := OX (X), F(f ) = f ∗ . We now describe
a quasi-inverse functor G from the affine algebras to the affine varieties
(this means that F ◦ G ∼ = Id and G ◦ F ∼ = Id, i.e. F and G estab-
lish an equivalence of categories, see Problem 4.6.1. In particular, if
(X, OX ), (Y, OY ) are affine varieties and f : X → Y is a map, then f is
a morphism if and only if f ∗ maps k[Y ] to k[X], and f : X → Y is an
isomorphism if and only if f ∗ is an isomorphism from k[Y ] to k[X].
So let A be an affine k-algebra. We define G(A) to be the affine variety
Specm A = (X, OX ), where
X := Homk-alg (A, k)
(which in view of Hilbert’s Nulstellensatz, can be identified with the set

of the maximal ideals of A, whence the name). Note that the elements
of a can be considered as k-valued functions on X via
f (x) := x(f ) (f ∈ A, x ∈ X = Homk-alg (A, k)).
Now consider the topology on X whose basis consists of all Xf := {x ∈

X | f (x) 6= 0} for f ∈ A. In order to define a structure sheaf on the
4.2 Prevarieties 49
topological space X, set
OX (Xf ) := Af (f ∈ A \ {0})
(again elements of Af can be considered as functions on Xf in a natural

way). Now for any U = ∪f Xf a function on U is in OX (U ) if and only
if its restriction to each Xf is in OX (Xf ).
Example 4.1.9 In view of Example 4.1.7, a closed subset of an affine

variety is an affine variety (as usual, with the induced sheaf), cf. Prob-
lem 4.6.6.
Example 4.1.10 If (X, OX ) is an affine variety, then it is easy to check

that the principal open set Xf is also an affine variety (think why this
does not contradict what was claimed in Example 4.1.7.) On the other
hand, not every open subset of X is an affine variety, see Problem 4.6.4.
4.2 Prevarieties
Definition 4.2.1 An (algebraic) prevariety is a geometric space (X, OX )

such that X has an open covering X = U1 ∪ · · · ∪ Ul , and each geometric
space (Ui , OUi ) with the induced structure sheaf OUi is an affine variety.
Example 4.2.2 In view of §3.9, each projective algebraic set with the
sheaf of regular functions is a prevariety. We will refer to varieties iso-
morphic to projective algebraic sets with sheaves of regular functions as
projective varieties.
Lemma 4.2.3 Let (X, OX ) be a prevariety with affine open covering

X = U1 ∪ · · · ∪ Ul .
(i) X is a noetherian topological space.
(ii) Any open subset U of X is again a prevariety.
(iii) Any closed subset Z of X is again a prevariety.
Proof (i) follows from the fact that each Ui is noetherian.

(ii) As U = ∪i (U ∩ Ui ), it suffices to prove that each U ∩ Ui has
an affine open covering. But U ∩ Ui is an open subset of an affine Ui ,
so it is a union of the principal open sets in Ui , which are affine by
Example 4.1.10.
50 Varieties
(iii) Z = ∪i (Z ∩ Ui ), and closed subsets Z ∩ Ui of affine varieties are
affine.
A subset of a topological space is called locally closed if it is an in-
tersection of an open set and a closed set. It follows from above that a
locally closed subset of a prevariety is again a prevariety. We will refer
to the locally closed subsets as subprevarieties.
Theorem 4.2.4 (Affine Criterion) Let X, Y be prevarieties, and ϕ :

X → Y be a map. Assume that there is an affine open covering Y =
∪i∈I Vi and an open covering X = ∪i∈I Ui such that
(i) ϕ(Ui ) ⊂ Vi for each i ∈ I;
(ii) f ◦ ϕ ∈ OX (Ui ) whenever f ∈ OY (Vi ).
Then ϕ is a morphism.
Proof An affine open covering of X induces that of each Ui . So, by

extending the index set if necessary we reduce to the case where Ui are
also affine. Now by assumption, ϕi := ϕ|Ui : Ui → Vi is a morphism of
affine varieties. In particular, ϕi is continuous, whence ϕ is continuous.
Let V ⊂ Y be an open subset, f ∈ OY (V ), and U := ϕ−1 (V ). By (ii),
f ◦ϕ ∈ OX (ϕ−1 (V ∩Vi )). But ϕ−1 (V ∩Vi ) ⊇ U ∩Ui , so f ◦ϕ ∈ OX (U ∩Ui )
for all i. Now, since U is the union of the U ∩ Ui and since OX is a sheaf,
f ◦ ϕ ∈ OX (U ).
Let X be an irreducible prevariety. Consider pairs (U, f ) where U is
an open subset of X and f ∈ OX (U ). We call two such pairs (U, f ) and
(U 0 , f 0 ) equivalent if there is a non-empty open subset V ⊂ U ∩ U 0 such
that f |V = f 0 |V (in which case we will also have f |(U ∩ U 0 ) = f 0 |(U ∩
U 0 )). It is easy to check using the irreducibility of X that this defines
an equivalence relation. Moreover, the set of equivalence classes is a
field with respect to the obvious operations. (For example, (U, f )−1 =
(U ∩ Uf , 1/f )). This field is called the field of rational functions on X
and denoted k(X). It is easy to see that if X is affine then this definition
agrees with the one we had before. Moreover, if U ⊂ X is a non-empty
open subset, then k(X) = k(U ).
Let F be a sheaf of functions on a topological space X and x ∈ X.
The open sets in X containing x form inverse system with respect to
inclusion. The stalk Fx of F at x is defined to be the corresponding
limit of algebras
Fx = lim F(U ).
U
4.3 Products 51
The elements of the stalk Fx are called germs of functions at x. One
can think of germs as equivalence classes of pairs (U, f ), where U is an
open set containing x, f ∈ F(U ), and (U, f ) ∼ (V, g) if there is an open
set W ⊂ U ∩ V containing x such that f |W = g|W . If (X, OX ) is a
prevariety, we write simply Ox for (OX )x and call it the local ring of x.
It is easy to see that the ring Ox is local in the sense of commutative
algebra. Its unique maximal ideal is denoted mx —it consists of the germs
of functions equal to zero at x.
If X is an irreducible affine variety, this definition agrees with the one
given in §3.7. Note also that Ox is a ‘local notion’, which means that if
x ∈ U for an open subset U ⊂ X, and OU is the induced sheaf on U ,
then Ox defined using U is the same as the one defined using X.
4.3 Products
Theorem 4.3.1 Finite products exist in the category of prevarieties.
Proof It suffices to deal with two prevarieties (X, OX ) and (Y, OY ).

We need to prove that there exists a prevariety (Z, OZ ) together with
morphisms π1 : Z → X and π2 : Z → Y such that the following universal
property holds: if (W, OW ) is another prevariety with morphisms ϕ1 :
W → X and ϕ2 : W → Y , then there exists a unique morphism ψ :
W → Z such that πi ψ = ϕi for i = 1, 2.
For any set S denote by Map(S, k) the algebra of all functions from
S to k. Observe that for any open U ⊂ X and V ⊂ Y the natural map
of algebras
OX (U ) ⊗ OY (V ) → Map(U × V, k).
is injective. So we will identify elements of OX (U ) ⊗ OY (V ) as functions

on U × V .
Now define a topology on the set X × Y by saying that the open sets
will be the unions of the sets of the form
(U × V )h := {x ∈ U × V | h(x) 6= 0},
where U ⊂ X, V ⊂ Y are arbitrary open subsets and h ∈ OX (U ) ⊗

OY (V ). We will refer to such (U × V )h as principal open sets. Checking
that this is a topology boils down to
(U × V )h ∩ (U 0 × V 0 )h0 = ((U ∩ U 0 ) × (V ∩ V 0 ))hh0 .

52 Varieties
Next we define a structure sheaf on X × Y . Let W be an open set in
X × Y and f ∈ Map(W, k). Then we say that f is regular if and only if
there is an open cover of W by the principal open sets (U × V )h so that
on each of them we have
a
f |(U × V )h =
hm
for some a ∈ OX (U ) ⊗ OY (V ) and some non-negative integer m.

This defines a sheaf OX×Y . Indeed, let W 0 ⊂ W be an open subset.
We have W = ∪(U × V )h and W 0 = ∪(U 0 × V 0 )h0 . So
W 0 = ∪((U × V )h ∩ (U 0 × V 0 )h0 ) = ∪((U ∩ U 0 ) × (V ∩ V 0 ))hh0 .
Moreover,
a 0 0 (a↓)h0m
|((U ∩ U ) × (V ∩ V ))hh0 = ,
hm (hh0 )m
where a↓ denotes the restriction of a from U × V to (U ∩ U 0 ) × (V ∩ V 0 ),

which belongs to OX (U ∩ U 0 ) ⊗ OY (V ∩ V 0 ). So f |W 0 is regular. The
second axiom of sheaf is obvious.
Now we want to show that (X × Y, OX×Y ) is a prevariety.
First, it is easy to see that for the case where X, Y are affine, our
definition agrees with the one from §3.6. So if X = ∪i Ui , Y = ∪j Vj are
open affine covers, then X × Y = ∪i,j Ui × Vj is an open affine cover.
Let π1 : X × Y → X and π2 : X × Y → Y be the natural projections,
and let us check the universal property. First of all, we need to check
that the projections are morphisms. They are continuous: for example,
for an open U ⊂ X, we have π −1 (U ) = U × V , which is open. Moreover,
let f ∈ OX (U ). Then (π1∗ (f ))(x, y) = f (x). So π1∗ (f ) = f ⊗ 1 ∈
OX (U ) ⊗ OY (Y ) is regular.
Finally, let ϕ1 : W → X and ϕ2 : W → Y be morphisms. It is clear
that if ψ required in the universal property exists, then it must send
w ∈ W to (ϕ1 (w), ϕ2 (w)). To show that ψ is a morphism, we use the
affine criterion. We know that the products U × V of the affine open
subsets cover X × Y . Open subsets of the form W 0 = ϕ−1 −1
1 (U ) ∩ ϕ2 (V )
∗ 0
P
cover W , and ψ maps a function ai ⊗ ai from OX×Y (U × W ) to
P ∗
the function ϕ1 (ai )ϕ∗2 (a0i ) ∈ OW (W 0 ). By the affine criterion, ψ is a
morphism.
4.4 Varieties 53
4.4 Varieties
Definition 4.4.1 A prevariety X is called an (algebraic) variety if the

diagonal ∆(X) = {(x, x) | x ∈ X} is closed in X × X.
An equivalent condition is as follows: for any prevariety Y and any
two morphisms ϕ, ψ : Y → X the set {y ∈ Y | ϕ(y) = ψ(y)} is closed in
Y . Indeed, applying this condition to π1 , π2 : X × X → X we conclude
that ∆ is closed; conversely, the preimage of ∆ under ϕ×ψ : Y → X ×X
is {y ∈ Y | ϕ(y) = ψ(y)}.
It follows from the previous paragraph that a subprevariety of a variety
is variety. We will refer to it a subvariety from now on.
In the category of topological spaces with usual product topology on
X × X the Definition 4.4.1 is equivalent to the Hausdorff axiom. So
we can think of varieties as prevarieties with some sort of an unusual
Hausdorff axiom.
Example 4.4.2 An example of a prevariety which is not a variety is

given by the affine line with a doubled point, see Problem 4.6.7.
Lemma 4.4.3 Let Y be a variety and X be a prevariety.

(i) If ϕ : X → Y is a morphism, then the graph
Γϕ := {(x, ϕ(x)) | x ∈ X}
is closed in X × Y .
(ii) If ϕ, ψ : X → Y are morphisms which agree on a dense subset of
X then ϕ = ψ.
Proof (i) Γϕ is the inverse image of ∆(Y ) with respect to the morphism
X × Y → Y × Y, (x, y) → (ϕ(x), y).
(ii) The set of all points where ϕ and ψ agree is closed.
Lemma 4.4.4 Affine varieties are varieties.
Proof Note that

∆(X) = {(x, y) ∈ X × X | evx = evy }
= {(x, y) ∈ X × X | f (x) = f (y) for all f ∈ k[X]}
= Z(f ⊗ 1 − 1 ⊗ f | f ∈ k[X]}.
54 Varieties
Lemma 4.4.5 The product of two varieties is a variety.
Proof Under the isomorphism (X × Y ) × (X × Y ) →

˜ (X × X) × (Y × Y ),
∆(X × Y ) maps to ∆(X) × ∆(Y ), which is closed.
Lemma 4.4.6 Let X be a prevariety. If every pair of points x, y ∈ X

lie in an open affine subset, then X is a variety.
Proof Let Y be a prevariety and ϕ, ψ : Y → X be morphisms. Set

Z := {y ∈ Y | ϕ(y) = ψ(y)}. In oprder to show that Z is closed,
let z ∈ Z̄, and x1 = ϕ(z), x2 = ψ(z). By assumption, x1 and x2 lie
in an open affine subset V of X. Then U := ϕ−1 (V ) ∩ ψ −1 (V ) is an
open neighborhood of z, which must have a non-trivial intersection with
Z. But Z ∩ U = {y ∈ U | ϕ0 (y) = ψ 0 (y)} where ϕ0 , ψ 0 : U → V are
restrictions of ϕ, ψ to U . As V is a variety, Z ∩ U is closed in U . So
U \ (Z ∩ U ) is open subset whose intersection with Z is empty. Hence
z ∈ Z.
It follows easily from Lemma 4.4.6 that projective varieties are vari-
eties, see Problem 4.6.16.
4.5 Dimension
Recall that we have assigned to every irreducible variety its field of
rational functions k(X). As k(X) is a finitely generated field extension
of k, it has a finite transcendence degree tr. degk k(X) over k. This
degree is called the dimension of X and denoted dim X. In general
dimension of X is defined as the maximum of the dimensions of its
irreducible components.
Example 4.5.1
(i) dim An = dim Pn = n.
(ii) Dimension of a finite set is 0. Conversely, if dim X = 0, then X
is finite. Indeed, let X be an irreducible affine variety X ⊂ An
of dimension 0. Let t1 , . . . , tn be coordinates on An considered
as functions on X. Then ti are algebraic over k, so can take only
finitely many values. So X is finite.
Example 4.5.2 Grassmann variety Gr (n) is covered by the open subsets

µi1 ...ir 6= 0, isomorphic to Ar(n−r) , so dim Gr (n) = r(n − r).
4.5 Dimension 55
Proposition 4.5.3 Let X and Y be irreducible varieties of dimensions
m and n, respectively. Then dim X × Y = m + n.
Proof We may assume that X and Y are affine. Let s1 , . . . , sp and

t1 , . . . , tq be generators of the algebras k[X] and k[Y ], respectively. Then
s1 , . . . , sp and t1 , . . . , tq generate the fields k(X) and k(Y ) over k, respec-
tively. So we can choose transcendence bases out of them. After renum-
bering, if necessary, transcendence bases are s1 , . . . , sm and t1 , . . . , tn .
Recall that k[X ×Y ] = k[X]⊗k[Y ]. Let us write si for si ⊗1 and tj for
1 ⊗ tj . As s1 , . . . , sp , t1 , . . . , tq generate k[X × Y ], they also generate the
field k(X × Y ) over k. Moreover, these generators depend algebraically
on s1 , . . . , sm , t1 , . . . , tn . So it suffices to prove that s1 , . . . , sm , t1 , . . . , tn
are algebraically independent.
Assume there is an algebraic dependence f (s1 , . . . , sm , t1 , . . . , tn ) = 0.
Then for each fixed x ∈ X the function f (s1 (x), . . . , sm (x), t1 , . . . , tn ) is
zero on Y . As t1 , . . . , tn are algebraically independent, all coefficients
g(s1 (x), . . . , sm (x)) of the polynomial f (s1 (x), . . . , sm (x), T1 , . . . , Tn ) ∈
k[T1 , . . . , Tn ] are zero. As x was arbitrary and s1 , . . . , sm are alge-
braically independent, it follows that the polynomial g(S1 , . . . , Sm ) ∈
k[S1 , . . . , Sm ] is zero. Hence f (S1 , . . . , Sm , T1 , . . . , Tn ) = 0.
Proposition 4.5.4 Let X be an irreducible variety and Y be a proper

closed subvariety. Then dim Y < dim X.
Proof We may assume that Y is irreducible and that X is affine, say of

dimension d. Let A = k[X], Ā = k[Y ]. Then Ā = A/P for some non-
zero prime ideal P of A. The transcendence bases of k(X) and k(Y )
can be found in A and Ā. Assume that dim Y ≥ d. Then we can choose
d algebraically independent elements ā1 , . . . , ād ∈ Ā. These elements
are cosets of some a1 , . . . , ad ∈ A which are of course also algebraically
independent. Let b ∈ P be a non-zero element. As dim X = d, there
must exist a non-trivial algebraic dependence f (b, a1 , . . . , ad ) = 0, where
f (T0 , T1 , . . . , Td ) ∈ k[T0 , T1 , . . . , Td ]. Since b 6= 0 we may assume that T0
does not appear in all monomials of the polynomial f , i.e. the polynomial
g(T1 , . . . , Tn ) = f (0, T1 , . . . , Tn ) is non-zero. But then g(ā1 , . . . , ād ) = 0,
giving a contradiction.
Corollary 4.5.5 Let X be an irreducible affine variety and Y is an

irreducible closed subvariety of codimension 1. Then Y is a component
of the variety Z(f ) for some f ∈ k[X].
56 Varieties
Proof By assumption Y 6= X, so there exists a non-zero function f ∈
k[X] with f |Y = 0. Then Y ⊆ Z(f ) ( X. Let Z be an irreducible
component of Z(f ) containing Y . By Proposition 4.5.4, dim Z < dim X.
So dim Z = dim Y , and by Proposition 4.5.4 again, Y = Z.
Lemma 4.5.6 If X is an irreducible affine variety for which k[X] is a

u.f.d., then every closed subvariety of codimension 1 has form Z(f ) for
some f ∈ k[X].
Proof Let Y be the subvariety, and Y1 , . . . , Yl be the components of Y .

Then I(Y ) = ∩I(Yi ). So, if we can prove that I(Yi ) = (fi ), then I(Y ) =
(f1 . . . fl ) (as the fi must be powers of different irreducible elements).
Thus we may suppose that Y is irreducible. Let P = I(Y ), a non-zero
prime ideal in k[X]. It therefore contains an irreducible element f . So
(f ) is a prime ideal contained in P . If (f ) ( P , then Y = Z(P ) (
Z (f ) ( X, which contradicts the assumption that codimension of Y
is 1, thanks to Proposition 4.5.4.
Remark 4.5.7 The statement of Lemma 4.5.6 fails if k[X] is not a

u.f.d. For example, let X = Z(T1 T4 − T2 T3 ) ⊂ A4 . It contains the
planes L and L0 given by the equations T2 = T4 = 0 and T1 = T3 = 0,
respectively. Clearly, L ∩ L0 = {(0, 0, 0, 0)}. We claim that L is not
Z(f ) for any f ∈ k[X]. Otherwise, Z(f |L0 ) = {(0, 0, 0, 0)}, which is
impossible, because it has codimension 2 in Z 0 .
If X is an affine variety and f ∈ k[X] is a non-invertible element, then

the zero set Z(f ) is called a hypersurface in X. If k[X] is a u.f.d., the
irreducible components of this hypersurface are precisely hypersurfaces
defined by the irreducible components of f .
Proposition 4.5.8 All irreducible components of a hypersurface in An

have codimension 1.
Proof It suffices to consider the zero set X of an irreducible polyno-

mial p(T1 , . . . , Tn ). We may assume that (say) Tn appears in p, as p
is non-scalar. Let ti := Ti |X. So k(X) = k(t1 , . . . , tn ). In view of
Proposition 4.5.4 it suffices to prove that t1 , . . . , tn−1 are algebraically
independent.
Assume that there is a non-trivial polynomial relation g(t1 , . . . , tn−1 ) =
4.5 Dimension 57
0, so the polynomial g(T1 , . . . , Tn−1 ) is zero on X. It follows that g is
divisible by p, which is impossible since Tn appears in p.
The proof of the following more general fact requires more powerful
commutative algebra:
Theorem 4.5.9 Let X be an irreducible affine variety, 0 6= f ∈ k[X]

be a non-invertible elemet, and Y be an irreducible component of Z(f ).
Then Y has codimension 1 in X.
Proof Let Y1 , . . . , Yl be the components of Z(f ) different from Y , and

P := I(Y ), Pi := I(Yi ) be the corresponding (prime) ideals in k[X]. As
the intersection of prime ideals is radical, it follows from the Nullstel-
lensatz that
p
(f ) = P ∩ P1 ∩ · · · ∩ Pl .
Note by the Nullstellensatz that P 6⊃ P1 ∩ · · · ∩ Pl . Take g ∈ P1 ∩ · · · ∩ Pl

with g 6∈ P . Note that Xg is an irreducible affine variety of the same
dimension as X, and, by the choice of g, Y ∩ Xg is the zero set of f
in Xg . On the other hand, Y ∩ Xg is a principal open sunset of Y , so
it suffices to prove that its codimension in Xg is 1.p So from the very
beginning we may assume that Y = Z(f ) and P = (f ).
Now, apply Noether’s Normalization Lemma 2.2.28 to the domain
R := k[X]: R is integral over some subring S isomorphic to k[T1 , . . . , Td ],
where d = dim X. Let E = k(X) and F be the field of fractions of S.
Then E/F is finite (generated by fnitely many algebraic elements). By
Corollary 2.2.27, the norm map NE/F takes values in S on elements of
R.
Denote NE/F (f ) =: f0 ∈ S. We claim that f0 ∈ P . Let irr (f, F ) =
xk + a1 xk−1 + · · · + ak ∈ S[x], see Lemma 2.2.26. By Lemma 2.2.25, f0
is ±amk for some m. Now f0 ∈ (f ) ⊆ P , in view of
0 = (f k + a1 f k−1 + · · · + ak )am−1
k
= f (f k−1 am−1
k + a1 f k−2 am−1
k + · · · + ak−1 am−1
k ) ± f0 .
Let Q be the radical of the ideal (f0 ) in S. Then Q ⊆ S ∩ P . We

claim that Q = S ∩ P . Indeed, let g ∈ S ∩ P . Since g ∈ P , we have
g l = f h for some l ∈ N and h ∈ R. Computing the norms, we get
g l[E:F ] = NE/F (f )NE/F (h) = f0 NE/F (h).
As NE/F (h) ∈ S, we deduce that g ∈ Q.

58 Varieties
We conclude that Q is a prime ideal in S. Since S is a UFD, it follows
that f0 is a power of an irreducible polynomial p in S, whence Q = (p).
Clearly p is not a scalar. Considering S as the algebra of regular func-
tions on Ad , we now conclude that Z(Q) is an irreducible hypersurface
of codimension 1, thanks to Proposition 4.5.8. So the transcendence
degree of the quotient field of S/Q over k is d − 1. On the other hand,
R is integral over S implies that R/P is integral over S/(P ∩ S) = S/Q.
So the quotient field of R/P also has transcendence degree d − 1 over k.
But the last quotient field is k(Y ), so dim Y = d − 1.
Corollary 4.5.10 Let X be an irreducible variety, U be an open subset of

X, and f ∈ OX (U ) be a non-invertible element. Then every irreducible
component of the zero set of f in U has codimension 1 in X.
Proof Let Y be an irreducible component of the zero set of f in U , and V

be an affine open subset in X contained in U with Y ∩V 6= ∅. Then using
Theorem 4.5.9, we have dim Y = dim(Y ∩ V ) = dim V − 1 = dim X − 1.
Corollary 4.5.11 Let X be an irreducible variety, and Y ⊆ X be an

irreducible closed subset of codimension r. Then there exist irreducible
closed subsets Yi of codimension 1 ≤ i ≤ r, such that Y = Yr ⊂ Yr−1 ⊂
· · · ⊂ Y1 .
Proof By passing to the affine open subset which intersects Y , we may

assume that X is affine. Apply induction on r. If r = 1, there is
nothing to prove. Since Y 6= X, there exists a function f 6= 0 in I(Y ),
and Y lies in an irreducible component Y1 of Z(f ). By Theorem 4.5.9,
codim Y1 = 1, and we can apply induction.
Corollary 4.5.12 (Topological Characterization of Dimension)

The dimension of an irreducible variety X is the largest integer d for
which there exist a chain of non-empty irreducible closed subsets
X0 ( X1 ( · · · ( Xd = X.
Proof This follows from Corollary 4.5.11 and the fact that the dimen-
sion of a proper closed subset of a variety is strictly smaller than the
dimension of the variety.
4.5 Dimension 59
Remark 4.5.13 The topological characterization shows that, when X
is irreducible affine, dim X is the Krull dimension dim k[X] of k[X], i.e.
the maximal length d of the chain of prime ideals 0 ( P0 ( P1 ( · · · (
Pd ( k[X]. Now Theorem 4.5.9 can be restated as follows: let A be
an affine k-algebra which is a domain, and f ∈ A be neither zero nor a
unit, and let P be a prime ideal minimal among those containing (f );
then dim A/P = dim A − 1. This statement is a version Krull’s principal
ideal theorem.
Corollary 4.5.14 Let X be an irreducible variety, f1 , . . . , fr ∈ OX (X).

Then each irreducible component of the set Z(f1 , . . . , fr ) has codimen-
sion at most r.
Proof Apply Theorem 4.5.9.
Remark 4.5.15 Let X = An , f1 = T1 , f2 = T1 +1. Then Z(f1 , f2 ) = ∅,

which is of codimension ∞, because by agreement dim ∅ = −∞. Think
why this does not contradict Corollary 4.5.14.
Corollary 4.5.16 Let X be an irreducible affine variety, and Y ⊂ X be

a closed irreducible subset of codimension r ≥ 1. Then Y is a component
of Z(f1 , . . . , fr ) for some f1 , . . . , fr ∈ k[X].
Proof We prove more generally that for closed irreducible subsets Y1 ⊃

Y2 ⊃ · · · ⊃ Yr with codim Yi = i there exist functions fi ∈ k[X] such
that all components of Z(f1 , . . . , fi ) have codimension i, and Yi is one of
those components (1 ≤ i ≤ r). This is indeed a more general statement
in view of Corollary 4.5.11.
Apply induction on i. For i = 1 we use Corollary 4.5.5 to find a func-
tion f1 such that Y1 is a component of Z(f1 ), and then Theorem 4.5.9
to deduce that all components of Z(f1 ) have codimension 1.
Assume that the functions f1 , . . . fi−1 have been found, and let Yi−1 =
Z1 , Z2 , . . . , Zm be the irreducible components of Z(f1 , . . . , fi−1 ). Each of
them has codimension i − 1, so none of them lies in Yi . So I(Zj ) 6⊃ I(Yi )
for all j = 1, . . . , m. The ideals I(Zj ) are prime, so it follows from
Theorem 2.1.5 that their union also does not contain in I(Yi ). Let fi be
a function which is zero on Yi but which is not identically zero on all Zj .
If Z is a component of Z(f1 , . . . , fi ), then Z lies in one of the com-
ponents Zj of the set Z(f1 , . . . , fi−1 ), and also in Z(fi ). So Z is a
component of Z(fi ) ∩ Zj , which by Theorem 4.5.9, has codimension 1 in
60 Varieties
Zj , and hence codimension i in X. Finally, the function fi is zero on Yi ,
and Yi has codimension i, so Yi is one of the components of Z(f1 , . . . , fi ).
Remark 4.5.17 The statement shows that for a prime ideal P in an

affine k-algebra which is a domain, if P has height r , then there ex-
ist elements f1 , . . . , fr such that P is minimal among the prime ideals
containing (f1 , . . . , fr ).
Remark 4.5.18 A closed subvariety X of An (resp. Pn ) of codi-

mension r is called a set theoretic complete intersection if there exist
r polynomials fi ∈ k[T1 , . . . , Tn ] (resp. r homogeneous polynomials
fi ∈ k[S0 , S1 , . . . , Sn ]) such that X = Z(f1 , . . . , fr ). Moreover, X is
called an ideal theoretic complete intersection if the fi can be chosen so
that I(X) = (f1 , . . . , fr ).
4.6 Problems
Problem 4.6.1 Prove that the functors F : (X, OX ) 7→ k[X] and G :

A 7→ Specm A are quasi-inverse equivalences of categories between affine
varieties over k and affine k-algebras (this means FG ∼
= Id and GF ∼
= Id).
Solution. To prove that FG ∼ = Id, let A be an affine k-algebra. By

definition, k[Specm A] ∼ = A, where A is considered as an algebra of
functions on Specm A via a(x) = x(a), see Remark 4.1.8. It is easy to
see that the isomorphism k[Specm A] ∼ = A is natural.
Now, let (X, OX ) be an affine variety. It follows from the axioms of
the affine variety and the definition of Specm k[X] that
X → Specm k[X], x 7→ evx
is an isomorphism of varieties, which is clearly natural. So GF ∼

= Id.
Problem 4.6.2 True or false? Let X be a prevariety and U ⊂ X is a

non-empty open subset. If f ∈ OX (U ) then f is a morphism from the
prevariety U to k = A1 .
Problem 4.6.3 Principal open sets in affine varieties are affine varieties.
Problem 4.6.4 Prove that A2 \ {(0, 0)} is not an affine variety.

4.6 Problems 61
Problem 4.6.5 Intersection of affine open subsets is affine.
Problem 4.6.6 Prove that a closed subset of an affine variety is again

an affine variety without using affine algebraic sets.
Problem 4.6.7 Make sense out of Example 4.4.2.
Problem 4.6.8 The product of irreducible prevarieties varieties is irre-

ducible.
Problem 4.6.9 Let ϕ1 : X1 → Y1 and ϕ2 : X2 → Y2 be morphisms

of prevarieties. Then ϕ1 × ϕ2 : X1 × X2 → Y1 × Y2 , (x1 , x2 ) 7→
(ϕ1 (x1 ), ϕ2 (x2 )) is also a morphism of prevarieties.
Problem 4.6.10 Let X, Y be prevarieties. Prove that the projections

X × Y → X and X × Y → Y are open maps, i.e. map open maps to
open maps. Do they have to map closed sets to closed sets?
Problem 4.6.11 Let ϕ : X → Y be prevarieties. Prove that the pro-

jection π1 induces an isomorphism from Γϕ ⊂ X × Y onto X.
Problem 4.6.12 Let X, Y be prevarieties, and X 0 ⊂ X, Y 0 ⊂ Y be sub-

prevarieties. Explain how X 0 × Y 0 can be considered as a subprevariety
of X × Y .
Problem 4.6.13 Prove that any morphism P1 → A1 must be constant.
Problem 4.6.14 Let f : A1 → A1 be a morphism. Then there is a

unique extension morphism f˜ : P1 → P1 such that f |A1 = f .
Problem 4.6.15 Show that every isomorphism f : P1 → P1 is of the

form f (x) = ax+b
cx+d for some a, b, c, d ∈ k, where x is the coordinate on
A1 .
Problem 4.6.16 Prove that Pn is a variety.
Problem 4.6.17 Prove that the Veronese embedding is an isomorphism

of Pn onto its image.
Problem 4.6.18 Let X ⊂ Pn be a projective algebraic set considered

as a variety and f ∈ k[S0 , . . . , Sn ] be a non-constant homogeneous poly-
62 Varieties
nomial. Then X \ Z(f ) is an affine variety. (Hint: Reduce to the case
where f is linear using the Veronese embedding).
Problem 4.6.19 Prove that the product of projective varieties defined

in §3.10 using Segre embedding is the categorical product.
Problem 4.6.20 Irreducible closed subvarieties of a variety X satisfy

A.C.C.
Problem 4.6.21 The dimension of a linear subvariety of An (that is a

subvariety defined by linear equations) has the value predicted by linear
algebra.
Problem 4.6.22 Let X and Y be closed subvarieties of An . For any non-

empty irreducible component Z of X ∩Y , we have codim Z ≤ codim X +
codim Y .
Problem 4.6.23 Fill in the details for Example 4.5.2
Problem 4.6.24 Prove that X × {point} ∼

= X.
5
Morphisms
5.1 Fibers
A fiber of a morphism ϕ : X → Y is a subset of the form ϕ−1 (y) for
y ∈ Y . As ϕ is continuous, fibers of ϕ are closed subvarieties in Y . Of
course ϕ−1 (y) is empty if y 6∈ im ϕ.
If X is irreducible and ϕ(X) is dense in Y we say that the morphism ϕ
is dominant. In this case Y will also have to be irreducible, as the image
of an irreducible topological space under a continuous map is irreducible
and the closure of an irreducible subspace is irreducible. More generally,
if X is not necessarily irreducible, then a morphism ϕ : X → Y is
dominant, if ϕ maps every component of X onto a dense subset of some
component of Y , and im ϕ is dense in Y .
If ϕ is a dominant morphism of irreducible varieties then the co-
morphism ϕ∗ induces an embedding of k(Y ) into k(X). In particular,
dim X ≥ dim Y .
Let ϕ : X → Y be a morphism, and W ⊆ Y be an irreducible closed
subset. If the restriction of ϕ to an irreducible component Z of ϕ−1 (W )
is dominant as a morphism from Z to W , then we say that Z dominates
W . If im ϕ ∩ W is dense in W then at least one of the components of
ϕ−1 (W ) dominates W .
Theorem 5.1.1 Let ϕ : X → Y be a dominant morphism of irreducible

varieties, and let r = dim X − dim Y . Let W be a closed irreducible
subset of Y , and Z be a component of ϕ−1 (W ) which dominates W .
Then dim Z ≥ dim W + r. In particular, if y ∈ im ϕ, then the dimension
of each component of the fiber ϕ−1 (y) is at least r.
Proof Let U be an affine open subset of Y which intersects W . Then
63
64 Morphisms
U ∩ W is dense in W , and hence irreducible. Also, ϕ(Z) ∩ U is dense
in W . So for the purpose of comparing dimensions we can consider U
instead of Y , W ∩ U instead of W , ϕ−1 (U ) ∩ Z instead of Z, and ϕ−1 (U )
instead of X. Thus we may assume that Y is affine.
Let s = codim Y W . By Corollary 4.5.16, W is an irreducible compo-
nent of Z(f1 , . . . , fs ) for some f1 , . . . , fs ∈ k[Y ]. Setting gi = ϕ∗ (fi ) ∈
OX (X), we have Z ⊆ Z(g1 , . . . , gs ). As Z is irreducible, it actually lies in
some component Z0 of Z(g1 , . . . , gs ). But by assumption W = ϕ(Z), and
ϕ(Z) ⊆ ϕ(Z0 ) ⊆ Z(f1 , . . . , fs ). As W is a component of Z(f1 , . . . , fs ), it
follows that ϕ(Z) = ϕ(Z0 ) = W , whence Z0 ⊆ ϕ−1 (W ). But Z is a com-
ponent of ϕ−1 (W ), so Z = Z0 , i.e. Z is a component of Z(g1 , . . . , gs ).
In view of Corollary 4.5.14, codim X Z ≤ s. The theorem follows.
The theorem says that the non-empty fibers of a morphism are not
‘too small’. The following example shows that they can be ‘too large’.
Example 5.1.2 Let ϕ : A2 → A2 be the morphism given by ϕ(x, y) =

(x, xy). Then ϕ is dominant. The fiber ϕ−1 ((0, 0)) is the y-axis, so it is
1-dimensional. On the other hand, all other non-empty fibers have the
‘right’ dimension 0.
5.2 Finite morphisms

Let ϕ : X → Y be a morphism of affine varieties. If the ring k[X]
is integral over the subring ϕ∗ (k[Y ]), then we say that the morphism
ϕ is finite. The main case is when X and Y are irreducible and ϕ is
dominant and finite. Then we can consider k[Y ] as a subring of k[X]
and then k(Y ) as a subfield of k(X). Moreover, since k[X] is integral
and finitely generated over k[Y ], k(X) is a finite algebraic extension of
k(Y ), so dim X = dim Y .
Fibers of finite maps are finite sets (which explains the terminology).
Indeed, let ϕ : X → Y be finite, X ⊂ An , and t1 , . . . , tn be the coor-
dinates on An as functions on X. By definition, each ti satisfies some
equation of the form tki + ϕ∗ (a1 )tk−1
i + · · · + ϕ∗ (ak ) = 0 with ai ∈ k[Y ].
−1
Let y ∈ Y and x ∈ ϕ (y). Then
ti (x)k + a1 (y)ti (x)k−1 + · · · + ak (y) = 0,
which has only finitely many roots.

Note that the morphism ϕ from Example 5.1.2 is dominant but not
finite. Indeed, T2 is not integral over ϕ∗ (k[Y ]) = k[T1 , T1 T2 ].
5.2 Finite morphisms 65
Remark 5.2.1 If ϕ : X → Y is a surjective morphism of irreducible
affine varieties, and all fibers are finite, then it can be proved that ϕ is
finite, see [Sp, §5.2]. We will not pursue this now.
Example 5.2.2 Let X be an affine variety and G be a finite group of

automorphisms of X, whose order N is prime to char k. We claim that
the projection map π : X → X/G is finite, cf. Example 3.5.7.
Theorem 5.2.3 Let ϕ : X → Y be a finite morphism of affine varieties

with f (X) dense in Y . Then ϕ(X) = Y .
Proof Let y ∈ Y , and let My be the corresponding maximal ideal of k[Y ].

If t1 , . . . , tn are the coordinate functions on Y and y = (a1 , . . . , an ), then
My = (t1 − a1 , . . . , tn − an ). Defining equations of the variety ϕ−1 (y)
are ϕ∗ (t1 ) = a1 , . . . , ϕ∗ (tn ) = an , and ϕ−1 (y) is empty if and only if
(ϕ∗ (t1 ) − a1 , . . . , ϕ∗ (tn ) − an ) = k[X]. If we identify k[Y ] with the
subring of k[X] via ϕ∗ , the last condition is equivalent to the condition
My k[X] = k[X].
Note that k[X] is a finitely generated k[Y ]-module in view of Propo-
sition 2.2.7(ii). So by Corollary 2.1.7, My k[X] 6= k[X].
Corollary 5.2.4 Finite maps are closed, i.e. they map closed sets onto
closed sets. To be more precise, let ϕ : X → Y be a finite map, and let
Z ⊂ X be a closed subset. Then ϕ|Z : Z → ϕ(Z) is finite. In particular,
ϕ(Z) = ϕ(Z).
Proof We may assume that ϕ(X) = Y . Denote R = k[X], S = k[Y ].

As ϕ∗ is injective, we can identify S with a subring of R, and then R is
integral over S, since ϕ is finite. If I is an ideal of R then R/I ⊃ S/(I ∩S)
is another integral ring extension.
Let I = I(Z). Then ϕ(Z) = Z 0 , where Z 0 = Z(I ∩ S). Moreover,
I := I ∩ S is radical, so I 0 = I(Z 0 ). The affine algebras of Z and
0
Z 0 are R/I and R/I 0 , so the remarks in the previous paragraph show
that ϕ|Z : Z → Z 0 is again finite and dominant. It remains to apply
Theorem 5.2.3 to this map.
Corollary 5.2.5 Let ϕ : X → Y be a finite dominant morphism of

irreducible affine varieties. Suppose that k[Y ] is integrally closed. If W
is a closed irreducible subset of Y and Z is any component of ϕ−1 (W ),
then ϕ(Z) = W .
66 Morphisms
Proof Keep the notation of the proof of Corollary 5.2.4, and let J =
I(W ). Then I ∩ S = I(ϕ(Z)) = I(ϕ(Z)) and I is a minimal prime ideal
of R for which I ∩S ⊇ J. It follows from the Going Down Theorem 2.2.29
that I ∩ S = J. So ϕ(Z) = W .
5.3 Image of a morphism

Let S ⊂ R be two finitely generated domains over k with quotient fields
E ⊂ F . Set r := tr. degE F . Let R0 be the localization of R with respect
to the multiplicative system S ∗ of non-zero elements of S. Note that F
is also field of fractions of R0 . On the other hand R0 contains E, so it can
be considered as an E-algebra. By Noether’s normalization lemma, R0
is integral over a ring E[T1 , . . . , Tr ] for some algebraically independent
elements T1 , . . . , Tr over E. Note that T1 , . . . , Tr can be chosen in R, as
all possible denominators are in E.
Now compare the integral extension E[T1 , . . . , Tr ] ⊂ R0 with the ex-
tension S[T1 , . . . , Tr ] ⊂ R. The latter extension is not necessarily inte-
gral, but R is finitely generated over S as a ring. Moreover, each gener-
ator of R satisfies a monic polynomial equation over E[T1 , . . . , Tr ]. If f
is a common denominator of all coefficients appearing in such equations
for all generators, then it is clear that Rf is integral over Sf [T1 , . . . , Tr ]
(and T1 , . . . , Tr are algebraically independent over Sf , because they are
algebraically independent even over E). These remarks will be used in
the proof of the following theorem.
Theorem 5.3.1 Let ϕ : X → Y be a dominant morphism of irreducible

varieties, and r = dim X − dim Y . Then
(i) im ϕ contains an open subset U of Y .
(ii) if all local rings of points of Y are integrally closed, then we
can choose U in part (i) so that it has the following property:
if W ⊂ Y is an irreducible closed subset which meets U , and Z
is a component of ϕ−1 (W ), which meets ϕ−1 (U ), then dim Z =
dim W + r.
Proof By passing to an open affine subset of Y , we may assume that

Y is affine (cf. the proof of Theorem 5.1.1). We may also reduce to the
case where X is affine. Indeed, let X = ∪Vi be an open affine covering.
As Vi is dense in X, we have ϕ(Vi ) is dense in X, so the restriction
ϕ|Vi : Vi → Y is a dominant morphism of irreducible affine varieties.
5.3 Image of a morphism 67
Now, if Ui is an open subset of Y as in (i) or (ii) for ϕ|Vi , then U = ∩i Vi
satisfies (i) and (ii), respectively.
Let R = k[X], S = k[Y ]. Consider S as a subring of R in a usual
way, and find elements T1 , . . . , Tr ∈ R, f ∈ S, such that Rf is integral
over Sf [T1 , . . . , Tr ]. Recall that Rf = k[Xf ] and Sf = k[Yf ]. So the
affine algebra Sf [T1 , . . . , Tr ] ∼
= Sf ⊗ k[T1 , . . . , Tr ] can be considered as
k[Yf × Ar ]. Then the restriction ϕ|Xf : Xf → Yf can be decomposed
ψ π
as a composition Xf −→ Yf × Ar −→ 1
Yf where ψ is a finite dominant
morphism. Set U = Yf and note that ϕ−1 (U ) = Xf . Moreover, ψ is
surjective by Theorem 5.2.3, and π1 is obviously surjective, so U ⊆ ϕ(X),
which proves (i).
To prove (ii), we also set X = Xf , U = Y = Yf . Then as above
ϕ = π1 ◦ ψ, where ψ is a finite morphism. It follows from the as-
sumption and (3.3) that the ring k[Y ] = Sf is integrally closed. Now
by Theorem 2.2.15, Sf [T1 , . . . , Tr ] is also integrally closed. If W is a
closed irreducible subset of Y and Z is any component of ϕ−1 (W ),
then Z is a component of ψ −1 (W × Ar ). Hence ψ(Z) = W × Ar , and
dim Z = dim ψ(Z) = dim W + r, see Corollary 5.2.5.
In (ii) above it would be enough to assume that local rings are inte-
grally closed only for some non-empty open subset of Y (we could pass
from Y to this open subset in the very beginning of the proof). It will
later turn out that this condition is always satisfied, see Theorem 6.3.1.
So the assumption can actually be dropped.
Proposition 5.3.2 Let ϕ : X → Y be a bijective morphism of irreducible

varieties. Then dim X = dim Y , and there are open subsets U ⊂ X and
V ⊂ Y such that ϕ(U ) = V and ϕ|U : U → V is a finite morphism.
Proof We may assume that Y is affine. Let W ⊂ X be an open affine

subset. As W is dense in X, we have ϕ(W ) is dense in Y , so the
restriction ϕ|W : W → Y is a dominant morphism of irreducible affine
varieties. Let R = k[W ], S = k[Y ]. Consider S as a subring of R
via (ϕ|W )∗ , and find elements x1 , . . . , xr ∈ R, f ∈ S, such that Rf is
integral over Sf [x1 , . . . , xr ]. Recall that Rf = k[Wf ] and Sf = k[Yf ]. So
the affine algebra Sf [x1 , . . . , xr ] ∼
= Sf ⊗ k[x1 , . . . , xr ] can be considered
as k[Yf × Ar ]. Then the restriction ϕ|Wf : Wf → Yf can be decomposed
ψ π
as a composition Wf −→ Yf × Ar −→ 1
Yf where ψ is a finite dominant
morphism. Now, ψ is surjective by Theorem 5.2.3. Hence ϕ|Wf : Wf →
68 Morphisms
Yf is surjective, and hence bijective by our assumption. This is only
possible if r = 0, so ϕ|Wf : Wf → Yf is finite, and dim X = dim Y .
Recall that a subset of a topological space is called locally closed if

it is an intersection of an open set and a closed set. A finite union of
locally closed sets is called a constructible set.
Theorem 5.3.3 Let ϕ : X → Y be a morphism of varieties. Then ϕ

maps constructible sets onto constructible sets. In particular, im ϕ is
constructible.
Proof Locally closed subset of a variety is itself a variety, so it suffices to

prove that im ϕ is constructible. We can also assume that X and Y are
irreducible. Apply induction on dim Y . If dim Y = 0, there is nothing
to prove. By inductive assumption, we may assume that ϕ is dominant.
Let U be an open subset contained in im ϕ, see Theorem 5.3.1(i).
Then the irreducible components Wi of Y \ U have dimensions less than
dim Y . By induction, the restriction of ϕ to Zi := ϕ−1 (Wi ) has image
constructible in Wi , so also constructible in Y . Now, ϕ(X) is a union of
U and the constructible sets ϕ(Zi ), so ϕ(X) is also constructible.
Proposition 5.3.4 Let ϕ : X → Y be a dominant morphism of irre-

ducible varieties.
(i) The set {y ∈ Y | dim ϕ−1 (y) ≥ n} is closed for any n.

(ii) For x ∈ X let εϕ (x) denote the maximal dimension of any com-
ponent of the set ϕ−1 (ϕ(x)) containing x. Then for all n ≥ 0,
the set En (ϕ) := {x ∈ X | εϕ (x) ≥ n} is closed in X.
Proof We prove (ii), the proof of (i) is very similar (and easier). Ap-
ply induction on dim Y , the case dim Y = 0 being clear. Let r =
dim X − dim Y , and let U be an open subset contained in im ϕ, see
Theorem 5.3.1(i). By Theorem 5.1.1, εϕ (x) ≥ r for all x, so En (ϕ) = X
for n ≤ r, in particular En (ϕ) is closed in this case. Let n > r. By
Theorem 5.3.1, En (ϕ) ⊂ X \ ϕ−1 (U ). Let Wi be the irreducible compo-
nents of the set Y \ U , Wij be the irreducible components of ϕ−1 (Wi )
and ϕij : Zij → Wi be the restriction of ϕ. Since dim Wi < dim Y , the
set En (ϕij ) is closed in Zij , and hence in X. But for n > r we have
En (ϕ) = ∪i,j En (ϕij ).
5.4 Open and birational morphisms 69
5.4 Open and birational morphisms
Example 5.1.2 shows that the image of an open set under a morphism
does not have to be open.
Theorem 5.4.1 Let ϕ : X → Y be a dominant morphism of irre-

ducible varieties, and r = dim X − dim Y . Assume that for each closed
irreducible subset W ⊂ Y all irreducible components of ϕ−1 (W ) have
dimension r + dim W . Then ϕ is open.
Proof Let y ∈ Y . By assumption, all irreducible components of ϕ−1 (y)

have dimension r. In particular, ϕ−1 (y) 6= ∅, whence ϕ is surjective.
Moreover, let W ⊂ Y be a closed irreducible subset and Z be an irre-
ducible component of ϕ−1 (W ). By assumption, dim Z = r + dim W .
Note that ϕ(Z) = W , as otherwise dim ϕ(Z) < dim W , and Z is an
irreducible component of ϕ−1 (ϕ(Z)), so we get a contradiction with our
assumptions.
Now, let U be an open subset of X, V = ϕ(U ), and y ∈ V . Then
y = ϕ(x) for some x ∈ U . It suffices to prove that y is in the interior
of V . Otherwise y ∈ Y \ V . By Theorem 5.3.3, V is constructible, so
Y \ V is also constructible. It follows that y lies in the closure of some
locally closed subset O ∩ C contained in Y \ V , where O is open and C
is closed. We may assume that C = O ∩ C. Moreover, we may assume
that C is irreducible, so O ∩ C is dense in C.
Now, each of the irreducible components of the set C 0 := ϕ−1 (C)
dominates C. So the set O0 := ϕ−1 (O) intersects each of the components
non-trivially. So O0 ∩C 0 is dense in C 0 . But the set O0 ∩C 0 = ϕ−1 (O ∩C)
lies in a closed subset X \ U , whence C 0 ⊂ X \ U . This contradicts the
fact that x ∈ C 0 .
Irreducible varieties X and Y are called birationally isomorphic, if

k(X) is k-isomorphic to k(Y ). A birationally isomorphic varieties do
not have to be isomorphic, for example A1 is birationally isomorphic to
P1 . On the other hand:
Proposition 5.4.2 Let X and Y be irreducible varieties. Then X and

Y are birationally isomorphic if and only if there exist non-empty open
subsets U ⊂ Y and V ⊂ X which are isomorphic.
Proof The ‘if-part’ is clear. In the other direction, let ϕ : k(Y ) → k(X)
be a k-isomorphism. We may assume that X and Y are affine. Let
70 Morphisms
f1 , . . . , fn generate the ring k[X] over k. Then for each i we can write
fi = ϕ(g i)
ϕ(h) (gi , h ∈ k[Y ]). So ϕ induces an isomorphism k[Y ]h →
˜ k[X]ϕ(h) .
So we may take U = Yh and V = Xϕ(h) .
A bijective morphism does not have to be an isomorphism. In fact its

topological behavior and the effect of its comorphism on functions can
be quite subtle. A typical example is the Frobenius map Fr : A1 → A1 .
But even in characteristic 0 one cannot assert that a bijective map is
an isomorphism, see Problem 5.5.4. However, Zariski’s Main Theorem
claims that a bijective birational morphism ϕ : X → Y of irreducible
varieties has to be an isomorphism if Y is smooth. (The smoothness will
be defined in the next chapter. We will not prove Zariski’s theorem).
Theorem 5.4.3 Let ϕ : X → Y be a dominant, injective morphism of

irreducible varieties. Then k(X) is a finite purely inseparable extension
of ϕ∗ k(Y ).
Proof See Humhreys, Theorem 4.6.
5.5 Problems
Problem 5.5.1 Give an example of a constructible set which is not

locally closed.
Problem 5.5.2 Prove that the following are equivalent descriptions of

the constructible sets in a topological space X:
(i) Constructible sets are finite disjoint union of locally closed sets.
(ii) Constructible sets are the sets expressible as
X \ (X2 \ (X3 \ · · · \ Xn )) . . . )
for a nested sequence X1 ⊃ X2 ⊃ X3 ⊃ · · · ⊃ Xn of closed

subsets.
(ii) The class of constructible sets of X is the smallest class including
open subsets and closed under the operations of finite intersec-
tions and complementation.
Problem 5.5.3 Prove that a constructible subset of a variety contains

a dense open subset of its closure.
5.5 Problems 71
Problem 5.5.4 Define a morphism ϕ : A1 → A2 by ϕ(x) = (x2 , x3 ).
Then X := im ϕ is closed in A2 and the morphism ϕ : A1 → X is
bijective, birational and homeomorphism, but it is not an isomorphism.
6
Tangent spaces
In this chapter, unless otherwise stated all varieties are assumed to be

irreducible.
6.1 Definition of tangent space

If X is the curve f (T1 , T2 ) = 0 in A2 then our ‘multivariable calculus
intuition’ tells us the tangent space to X at x = (x1 , x2 ) ∈ X is the set
of solutions of the linear equation
∂f ∂f
(x)(T1 − x1 ) + (x)(T2 − x2 ) = 0.
∂T1 ∂T2
This ‘tangent space’ is a line unless both partial derivatives are zero at
x. More generally, if f ∈ k[T1 , . . . , Tn ] set
n
X ∂f
dx f = (x)(Ti − xi ).
i=1
∂T i
Now, if X ⊂ An is a closed subset and I = I(X) we define the geometric

tangent space Tan(X)x to X at x to be the linear variety Z(J) ⊂ An
where the ideal J is generated by all dx f for f ∈ I. We consider Tan(X)x
as a verctor space with the origin at x. Problem 6.8.1 is handy for explicit
calculations of geometric tangent spaces.
For any f (T ) ∈ k[T ], dx f can be considered as a linear function on
An with the origin at x, so on restriction to Tan(X)x , dx f is a linear
function on Tan(X)x . By definition, dx f = 0 on Tan(X)x for f ∈ I(X),
so we can define the linear function dx f on Tan(X)x for f ∈ k[X]. Thus
dx becomes a linear map from k[X] to Tan(X)∗x . It is surjective, as
any g ∈ Tan(X)∗x is the restriction of a linear polynomial f on An (as
usual, origin at x), and dx f = f . Let M be the maximal ideal of k[X]
72
6.1 Definition of tangent space 73
corresponding to x. As k[X] = k ⊕M , and dx maps constants to zero, dx
induces a surjective map from M to Tan(X)∗x . We claim that the kernel
of this map is M 2 . By the product rule, M 2 ⊆ ker dx . Conversely, let
f ∈ M and dx f = 0 on Tan(X)x . Assume that f is the image of some
polynomial function f (T ) on An . By Problem 6.8.1 and linear algebra,
P
we have dx f = i ai dx fi for some ai ∈ k and fi ∈ I(X). Then for
g := f − i ai fi we have dx g = 0 on An , which means that g does not
P
contain linear terms (Ti − xi ), i.e. g belongs to the square of the ideal
generated by all Ti − xi . The image of this ideal in k[X] is M , and the
image of g is f , so f ∈ M 2 .
Thus, we have identified the vector space Tan(X)∗x with M/M 2 or
Tan(X)x with (M/M 2 )∗ . Now, in view of Lemma 2.1.11, the vector
space M/M 2 can be identified with mx /m2x , where mx is the maximal
ideal of the local ring Ox . So, we have motivated the following ‘invariant’
definition.
Definition 6.1.1 The tangent space to the variety X at x ∈ X, denoted

Tx X is the k-vector space (mx /m2x )∗ .
We now give another description of Tx X. A derivation at x is a k-
linear map δ : Ox → k such that δ(f g) = δ(f )g(x) + f (x)δ(g). We claim
that the vector space of derivations at x is naturally isomorphic to Tx X.
Indeed, if δ : Ox → k is a derivation, it follows easily that δ(f ) = 0 if f
is a constant or if f ∈ m2x . So δ defines an element of (mx /m2x )∗ . This
defines a map from the space of derivations at x to Tx X, which is easily
shown to be an isomorphism.
Let X be an irreducible affine variety. We claim that in this case we
can also identify Tx X with the derivations of k[X] at x, i.e. the linear
maps δ : k[X] → k such that δ(f g) = δ(f )g(x) + f (x)δ(g). Indeed,
recall that under our assumptions Ox can be identified with the subring
of k(X) consisting of all rational functions which are regular at x. Now,
if δ : Ox → k is a derivation, we get a derivation δ̄ : k[X] → k on
restriction. Conversely, if δ : k[X] → k is a derivation and h = fg ∈ Ox
δ(f )g(x)−f (x)δ(g)
define δ̂(h) = g(x)2 (the ‘quotient rule’). It is easy to check
that the maps δ 7→ δ̄ and δ 7→ δ̂ are inverse to each other.
Example 6.1.2 Let X = An . The map ∂T ∂

i
|x : k[X] → k, f 7→ ∂f∂T(x)
i
is a derivation of k[X] = k[T1 , . . . , Tn ] at x. It is easy to check that
the derivations ∂T∂
1
|x , . . . , ∂T∂ n |x form a basis of Tx X, so Tx X ∼= k n . It
n ∼ n n
follows that Tx P = k for any x ∈ P .
74 Tangent spaces
Example 6.1.3 Let X ⊂ An be an affine irreducible variety. Then any
derivation δ of k[X] = k[T1 , . . . , Tn ]/I(X) can be lifted to a deriva-
tion δ̂ of k[X] at x. So by Example 6.1.2 any derivation of X at
Pn ∂
x looks like i=1 ai ∂Ti |x for some constants ai ∈ k. Moreover, if
Pn ∂
I(X) = (f1 , . . . , fl ), then i=1 ai ∂T i
|x is zero on I(X) if and only if
n
X ∂fj (x)
ai (j = 1, . . . , l). (6.1)
i=1
∂Ti
So Tx X is a linear space of all tuples (a1 , . . . , an ) ∈ k n satisfying the

equations (6.1).
Example 6.1.4 Let X be given by the equation y 2 = x3 . Then

I(X) = (y 2 − x3 ), and X is one-dimensional. On the other hand, using
Example 6.1.3, one sees that the tangent space Tx X is one-dimensional
for all x, except for x = (0, 0) when Tx X is two-dimensional. Soon we
will see that in general dim Tx X ≥ dim X, and that the equality holds
for ‘almost all’ points x ∈ X.
Proposition 6.1.5 Let X, Y be irreducible varieties, x ∈ X, y ∈ Y .

Then T(x,y) (X × Y ) ∼
= Tx X ⊕ Ty Y .
Proof We may assume that X and Y are affine. If δ1 : k[X] → k is a

derivation at x and δ2 : k[Y ] → k is a derivation of k[Y ] at y, define the
derivation
(δ1 , δ2 ) : k[X × Y ] = k[X] ⊗ k[Y ] → k, f ⊗ g 7→ δ1 (f )g(y) + f (x)δ2 (y)
of k[X × Y ] at (x, y). This defines an isomorphism from Tx X ⊕ Ty Y to

T(x,y) (X × Y ) (check!).
6.2 Simple points
Definition 6.2.1 Let X be an irreducible variety and x ∈ X. Then

x is called a simple point if dim Tx X = dim X. Otherwise x is called
singular. If all points of X are simple, then X is called smooth (or
non-singular).
So An and Pn are smooth and the product of smooth varieties is

smooth.
6.2 Simple points 75
Lemma 6.2.2 Let X = Z(f ) be an irreducible hypersurface in An . Then
dim Tx X = dim X for all points x from some open dense subset of X.
Proof We may assume that f is an irreducible polynomial. The tangent

space Tx X is the set of all n-tuples (a1 , . . . , an ) ∈ k n satisfying the linear
Pn
equation i=1 ai ∂f∂T(x)
i
= 0. Since dim X = n − 1, a point x is singular
if and only if all ∂f∂T(x)
i
∂f
= 0. If the polynomial ∂Ti
is non-zero, then it is
not identically zero on X, as otherwise it would be divisible by f , which
is impossible by degrees. So we may assume that char k = p and all
degrees of all variables Ti in f are divisible by p, but then f = g p by
‘Freshman’s Dream’, which contradicts the irreducibility of f .
Theorem 6.2.3 Let X be an irreducible variety. Then dim Tx X ≥

dim X for any x ∈ X, and the equality holds for all points x from some
open dense subset of X.
Proof By Theorem 2.1.12, k(X) is separably generated over k, i.e. k(X)

is a finite separable extension of L = k(t1 , . . . , td ), which in turn is
a purely transcendental extension of k. Note that d = dim X. By
the Primitive Element Theorem 2.1.13, K = L(t0 ) for some element
t0 ∈ K. Let f (T0 ) := irr (t0 ; L) ∈ L[T0 ] be the minimal polynomial.
Since the coefficients of f are rational functions in k(t1 , . . . , td ), this
polynomial can be considered as a rational function f (T0 , T1 , . . . , Td ) ∈
k(T0 , T1 , . . . , Td ). This rational function is defined on a principal open
subset of Ad+1 , and the zero locus Y of f is an irreducible hypersurface
in this principal open subset.
We claim that k(Y ) ∼ = k(X). Indeed, let si be the restriction of the
coordinate function Ti to Y for 0 ≤ i ≤ n. Then k(Y ) = k(s0 , s1 , . . . , sd ).
As dim Y = d and s0 is algebraic over k(s1 , . . . , sd ), we conclude that
s1 , . . . , sd are algebraically independent over k. Now, it is clear that the
minimal polynomial of s0 over k(s1 , . . . , sd ) is f , whence the claim.
By Proposition 5.4.2, there exist non-empty open subsets in X and
Y which are isomorphic. By Lemma 6.2.2, the set of points y ∈ Y for
which dim Ty Y = dim Y form an open subset in Y , so the same follows
for X.
Let x be an arbitrary point of X. In order to find the dimension of Tx X
we may pass to an affine open neighborhood of x. So we may assume that
X is a closed subset of some An . Then Tx X can be considered as a vector
subspace of k n . By shifting the origin to x we have Tx X as an affine
76 Tangent spaces
subspace of An through x. Let T be the subset of all (x, y) ∈ X × An for
which y ∈ Tx X. Note that T is a closed subset. Indeed, it is given by
the equations for X together with the polynomial equations of the form
Pn ∂fj (x) n
i=1 ∂Ti (Si − xi ), where Si are the coordinates in A . Projection
−1
pr1 defines a morphism ϕ : T → X whose fiber ϕ (x) has dimension
dim Tx X. For each m the subset Xm = {x ∈ X | dim ϕ−1 (x) ≥ m} is
closed in X, see Proposition 5.3.4(i). But we saw that Xd is dense in X,
so Xd = X.
6.3 Local ring of a simple point

Let X be an irreducible variety and x ∈ X. By Corollary 2.1.10, the
minimal number n of generators of the ideal mx equals the dimension of
mx /m2x or dim Tx X. So the point x is simple if and only if n = dim X.
Recall that the Krull dimension of a Noetherian ring R is defined to be
the largest length k of a chain
0 ( P1 ( P2 ( · · · ( Pk ( R
of prime ideals.
We claim that the Krull dimension of Ox equals dim X. Indeed, we
may assume that X is affine, in which case Ox = k[X]Mx . But the prime
ideals of k[X]Mx are in one-to-one correspondence with prime ideals of
k[X] contained in Mx , and it follows from Corollaries 4.5.11 and 4.5.12
that dim X is the largest length k of a chain
0 ( P1 ( P2 ( · · · ( Pk = Mx
of prime ideals in k[X].
A local ring (R, M ) is called regular if its Krull dimension equals the
number of generators of the maximal ideal M . We have established that
a point x ∈ X is simple if and only if its local ring Ox is regular. So, in
view of Theorem 2.1.15, we have:
Theorem 6.3.1 Let x be a simple point of an irreducible variety X.

Then Ox is a regular local ring. In particular, it is a UFD and is inte-
grally closed.
Theorem 6.3.2 Let X be an irreducible variety and x ∈ X be a point

such that Ox is integrally closed. Let f ∈ K(X) \ Ox . Then there exists
a subvariety Y ⊂ X containing x and such that f 0 := f1 ∈ Oy for some
y ∈ Y , and f 0 is equal to zero on Y everywhere where it is defined.
6.4 Differential of a morphism 77
Proof Let R = Ox . Then I := {g ∈ R | gf ∈ R} is a proper ideal
of R, as 1 6∈ I, and so I ⊂ mx . Let P = P1 , P2 , . . . , Pt be the distinct
minimal prime ideals containing I. Then P1 ∩ · · · ∩ Pt /I is nilpotent, i.e.
P n P2n . . . Ptn ⊂ I. For i > 1, Pi generates in the local ring RP ⊂ k(X)
the ideal coinciding with the whole RP . So P n RP ⊂ IRP . In particular,
since If ⊂ R, we have P n f ⊂ (If )Rp ⊂ RP . Choose k ≥ 0 the minimal
possible so that P k f ⊂ RP , and let g ∈ P k−1 f \ RP . Then P g ⊂ RP .
By assumption R is integrally closed, so RP is integrally closed, see
Proposition 2.2.13. As g 6∈ RP , the element g is not integral over
RP . Now, if P RP g ⊆ P RP , then the ring RP [g] acts faithfully on the
finitely generated RP -module P RP , giving a contradiction, see Proposi-
tion 2.2.5(iii). So P g ⊂ RP generates the ideal RP in RP , hence contains
an invertible element from RP . So g1 ∈ P RP , and P RP = g1 RP .
f
Now, h := gk
∈ f P k RP ⊂ RP . We claim that h is a unit in RP .
f
Otherwise h ∈ P RP = g1 RP or gk−1 ∈ RP , which contradicts the choice
1 −1 1
of k. So f = h gk ∈ P RP .
Let P be generated by the elements f1 , . . . , fl ∈ Ox . Then fi are
rational functions regular in some neighborhood of x, so also in some
affine neighborhood of x. Now let Y be the zero locus of the functions
f1 , . . . , fl in this affine neighborhood. Then all functions of P are zero
everywhere on Y where they are defined. So this is also true for the
function f1 ∈ P RP . Also x ∈ Y , as P ⊂ mx .
6.4 Differential of a morphism

Let ϕ : X → Y be a morphism of (irreducible) varieties, x ∈ X, y =
ϕ(x). Then ϕ∗ (Oy ) ⊂ Ox , and ϕ∗ (my ) ⊂ mx . So ϕ∗ induces a map
my /m2y → mx /m2x , which in turn induces a map (mx /m2x )∗ → (my /m2y )∗ .
This map is denoted dϕx and is called the differential of ϕ at x. Thus:
dϕx : Tx X → Tϕ(x) Y.
In terms of derivations, dϕx can be described similarly: if δ : Ox → k is
a derivation, then dϕx (δ) is defined to be δ ◦ ϕ∗ : Oy → k. The following
natural properties are easy to check:
dx id = id and d(ψ ◦ ϕ)x = dψϕ(x) ◦ dϕx .
Example 6.4.1 Let X ⊂ An and Y ⊂ Am be affine algebraic sets

and ϕ : X → Y be the restriction of ϕ = (ϕ1 , . . . , ϕm ) with ϕi ∈
k[T1 , . . . , Tn ]. Take x ∈ X and let y = ϕ(x). We identify Tx X and Ty Y
78 Tangent spaces
with subspaces of k n and k m , respectively, following Example 6.1.3. If
a = (a1 , . . . , an ) ∈ Tx X, then dϕx (a) = (b1 , . . . , bm ), where
X ∂ϕj
bj = (x)ai ,
i
∂Ti
i.e. dϕx is the linear map whose matrix is the Jacobian of ϕ at x.
Example 6.4.2 Let X = GLn (k), Y = GL1 (k) = k × , and ϕ = det.

2
Note that X is the principal open set in An which we identify with
Mn (k), all n×n matrices. It is easy to see that at every point x ∈ GLn (k)
the tangent space Tx GLn (k) can be identified with Mn (k). Let e be the
identity matrix. Under our identification, d dete : Mn (k) → M1 (k) = k
is tr, the trace map.
Let ϕ : X → Y be a dominant morphism of irreducible varieties. Then

k(Y ) can be considered as a subfield of k(X) via ϕ∗ . If the extension
k(X)/k(Y ) is separable, we say that the morphism ϕ is separable.
In characteristic 0 all morphisms are separable. An example of a non-
separable morphism is given by the Frobenius morphism.
We are going to develop some machinery which will help us to establish
a differential criterion for separability and to consider tangent spaces
from a new point of view.
6.5 Module of differentials

For a k-algebra A and an A-module M , we write
Derk (A, M )
for the space of all k-linear derivations from A to M , i.e. k-linear maps
f : A → M such that f (ab) = af (b) + bf (a) for all a, b ∈ A.
Let m : A ⊗k A → A be the multiplication, and let I := ker m, the
ideal generated by all a ⊗ 1 − 1 ⊗ a (a ∈ A). Define the module of
differentials ΩA/k to be
ΩA/k := I/I 2 .
This is an A ⊗k A-module annihilated by I, so it can be considered as a

module over A ∼= (A ⊗ A)/I. Let da denote the image of a ⊗ 1 − 1 ⊗ a
in ΩA/k . Note that the map d : a 7→ da is a derivation from A to the
A-module ΩA/k :
ad(b) + d(a)b = a(b ⊗ 1 − 1 ⊗ b) + (a ⊗ 1 − 1 ⊗ a)b + I 2

= ab ⊗ 1 − 1 ⊗ ab + I 2 = d(ab).
The elements da for a ∈ A generate ΩA/k as an A-module. One should

think of ΩA/k as the universal module for derivations of A:
Theorem 6.5.1 Suppose that M is an A-module and D : A → M is

a k-derivation. Then there exists a unique A-module homomorphism
ϕ : ΩA/k → M such that D = ϕ ◦ d, i.e. the map
HomA (ΩA/k , M ) → Derk (A, M ), ϕ 7→ ϕ ◦ d
is an isomorphism.
Proof Define the linear map
ψ : A ⊗ A → M, a ⊗ b 7→ bD(a).
One checks that for arbitrary elements x, y ∈ A ⊗ A,
ψ(xy) = m(x)ψ(y) + m(y)ψ(x),
hence ψ vanishes on I 2 . Therefore it induces a map ϕ : ΩA/k → M which

is actually an A-module map, such that ϕ(da) = ψ(a ⊗ 1 − 1 ⊗ a) = D(a)
(here we have used that D(1) = 0). For uniqueness use the fact that the
da generate ΩA/k as an A-module.
The theorem gives a universal property for the pair (ΩA/k , d) which
as usual characterizes it up to a unique A-module isomorphism.
Example 6.5.2 (i) Let F be any field, A = F [T1 , . . . , Tn ]/(f1 , . . . , fm ),

and ti = Ti + (f1 , . . . , fm ) ∈ A. Then the dti generate ΩA/F as an A-
module, since the ti generate A as an algebra. Moreover, the kernel of
the A-module homomorphism
n
M
An = Aei → ΩA/F , ei 7→ dti
i=1
is the submodule K of An generated by the elements

n
X ∂fj
(t1 , . . . , tn )ei (1 ≤ j ≤ m).
i=1
∂Ti
80 Tangent spaces
Indeed, consider the map
n
X ∂f
d0 : A → An /K, f 7→ (t1 , . . . , tn )ei
i=1
∂T i
∂f
( ∂T i
(t1 , . . . , tn ) means: take any representative f˜(T1 , . . . , Tn ) in F [T ],
take the partial derivative of f˜, and pass to the quotient again.) The
result follows from the fact that (An /K, d0 ) satisfy the universal property
of the theorem.
(ii) Consider two special cases of (i): when A = k[T1 , . . . , Tn ], then
ΩA/k is a free module on the basis dT1 , . . . , dTn ; when A = k[T1 , T2 ]/(T12 −
T23 ), then ΩA/k = (Ae1 ⊕ Ae2 )/(2t1 e1 − 3t22 e2 ), which is not a free A-
module.
(iii) Let A be an integral domain with quotient field E. Then ΩE/k =
E ⊗A ΩA/k . Indeed, the derivation d : A → ΩA/k induces a derivation
dˆ : E → E ⊗A ΩA/k . We claim that E ⊗A ΩA/k together with dˆ has
the correct universal property. Take an E-module M and a derivation
D̂ : E → M . Its restriction D to A is a derivation A → M . Hence
there exists a unique A-module homomorphism ϕ : ΩA/k → M with
D = ϕ ◦ d. Hence since M is an E-module, there is a unique E-module
homomorphism ϕ̂ : E ⊗A ΩA/k → M with D̂ = ϕ̂ ◦ d. ˆ
(iv) Suppose that E = k(x1 , . . . , xn ) is a finitely generated field ex-
tension of k. By (iii) and (i), ΩE/k is the E-vector space spanned by
dx1 , . . . , dxn . In particular, it is finite dimensional.
Example 6.5.3 Let X be an affine variety, x ∈ X, and kx = k be the

1-dimensional k[X]-module with action f · c = f (x)c for f ∈ k[X], c ∈ k.
Denote ΩX := Ωk[X]/k . By the theorem,
Homk[X] (ΩX , kx ) ∼
= Derk (k[X], kx ) ∼
= Tx X
Now, if X ⊂ An is closed and I(X) = (f1 , . . . , fm ), it follows from
Example 6.5.2(i) that ΩX is generated by dt1 , . . . , dtm and the relations
n
X ∂fj
(t1 , . . . , tn )dti = 0 (1 ≤ j ≤ m).
i=1
∂Ti
Now, it is clear that Homk[X] (ΩX , kx ) is the vector space of all n-tuples
(a1 , . . . , an ) satisfying equations
n
X ∂fj
(x)ai = 0 (1 ≤ j ≤ m).
i=1
∂Ti
So we recover the description of the tangent space from Example 6.1.3.
Now for the remainder of the section we will be concerned with the fol-
lowing situation: we are given finitely generated field extensions F/E/k.
Then there exists an exact sequence
0 → DerE (F, F ) → Derk (F, F ) → Derk (E, F ).
The first map is the obvious inclusion and the second map is induced by
restriction of functions from F to E. To check the exactness in the second
term, note that any D ∈ DerE (F, F ) maps E to zero, and conversely,
any f ∈ Derk (F, F ) that maps elements of E to zero is E-linear.
Applying the universal property we have an exact sequence
0 → HomF (ΩF/E , F ) → HomF (ΩF/k , F ) → HomE (ΩE/k , F ).
Also note that HomE (ΩE/k , F ) ∼

= HomF (F ⊗ ΩE/k , F ). So we have an
exact sequence of finite dimensional F -vector spaces
0 → HomF (ΩF/E , F ) → HomF (ΩF/k , F ) → HomF (F ⊗ ΩE/k , F ).
Dualizing we get
α β
F ⊗ ΩE/k −→ ΩF/k −→ ΩF/E −→ 0,
where α sends 1 ⊗ dE/k a to dF/k a, viewing a ∈ E as an element of F ,

and β is induced by the derivation dF/E : F → ΩF/E according to the
universal property of ΩF/k .
Lemma 6.5.4 If F is a finite dimensional separable extension of E then

α is injective.
Proof By the above discussion, this is equivalent to the restriction

map Derk (F, F ) → Derk (E, F ) being surjective. Equivalently, every k-
derivation from E to F can be extended to a derivation from F to F . By
the Primitive Element Theorem, we may assume that F = E[T ]/(f (T )),
where
f (T ) = T n + an−1 T n−1 + · · · + a0
is an irreducible polynomial and (this is what separability means) f 0 (x) 6=

0, where x is the image under the quotient map of T in F .
Let D : E → F be a derivation. To extend D to a derivation D̂ from
F to F , we just need to decide what D̂(x) should be: then the derivation
formula means that there is no choice for defining D̂ applied to any other
82 Tangent spaces
element of F = E[x]. To decide on D̂(x) we need for well-definedness
that D̂(f (x)) = 0, i.e.
X
f 0 (x)D̂(x) + D(ai )xi = 0.
Since f 0 (x) 6= 0, we can solve this equation for D̂(x) in the field F .
Lemma 6.5.5 Let F = E(x). Then dimF ΩF/E ≤ 1. Moreover, ΩF/E =

0 if and only if F/E is a finite separable extension.
Proof By Example 6.5.2(iii), ΩF/E = F ⊗E[x] ΩE[x]/E . If x is transcen-

dental over E, we have ΩE[x]/E is free of rank 1, cf. Example 6.5.2(i). If x
is algebraic, by Example 6.5.2(i) again we have ΩE[x]/E = E[x]/(f 0 (x)),
where f (x) = irr (x; E). If F/E is not separable, then f 0 (x) = 0, and
we again get that ΩF/E is one-dimensional. Finally, if F/E is separable,
then f 0 (x) 6= 0, and ΩF/E = F ⊗E[x] ΩE[x]/E = 0, since it is generated
by 1 ⊗ 1 = f 0 (x)−1 f 0 (x) ⊗ 1 = 0.
Theorem 6.5.6 (Differential Criterion for Separability) Let F =

E(x1 , . . . , xm ) be a finitely generated field extension. Then:
(i) dimF ΩF/E ≥ tr. degE F .
(ii) Equality in (i) holds if and only if F/E is a separable extension.
Proof Proceed by induction on d = dimF ΩF/E . If d = 0, i.e. ΩF/E = 0

to get (i) and (ii), we just need to show that F/E is a finite separa-
ble extension. For this we use induction on m, the case m = 1 be-
ing Lemma 6.5.5. Now suppose m > 1. Set E 0 = E(xm ), so F =
E 0 (x1 , . . . , xm−1 ). Using the exact sequence
α β
F ⊗ ΩE 0 /E −→ ΩF/E −→ ΩF/E 0 −→ 0,
we see that ΩF/E 0 = 0. Hence by induction F/E 0 is a finite separable

extension. So by Lemma 6.5.4, α is injective, whence ΩE 0 /E = 0, and
E 0 /E is a finite separable extension. By transitivity, F/E is a finite
separable extension.
Now suppose d > 0. Pick x ∈ F with dF/E x 6= 0, and let E 0 := E(x).
We have the exact sequence
α β
F ⊗ ΩE 0 /E −→ ΩF/E −→ ΩF/E 0 −→ 0.
Since α(1⊗dE 0 /E x) = dF/E x 6= 0, we have ΩE 0 /E 6= 0. So by Lemma 6.5.5,

6.6 Simple points revisited 83
dimE 0 ΩE 0 /E = 1, which means that α is injective. So dimF ΩF/E =
dimF ΩF/E 0 + 1. By induction, dimF ΩF/E ≥ tr. degE 0 F + 1. Since
tr. degE F = tr. degE 0 F + tr. degE E 0 ≤ tr. degE 0 F + 1,
we get dimF ΩF/E ≥ tr. degE F , which is (i). With a little further
argument along the same lines, one gets (ii).
Corollary 6.5.7 Assume that E ⊂ F are finitely generated field ex-

tensions of k. Then F/E is separable if and only if the natural map
Derk (F, F ) → Derk (E, F ) is surjective.
Proof As above, Derk (F, F ) → Derk (E, F ) is surjective if and only if

the map
α : F ⊗E ΩE/k → ΩF/k
is injective. Consider the exact sequence

α β
F ⊗ ΩE/k −→ ΩF/k −→ ΩF/E −→ 0.
As k is algebraically closed, every extension of k is separable, so by the

theorem, dimF F ⊗E ΩE/k = dimE ΩE/k = tr. degk E and dimF ΩF/k =
tr. degk F . Hence α is injective if and only if
dimF ΩF/E = tr. degk F − tr. degk E = tr. degE F.
By the theorem, this is if and only if F/E is separable.
6.6 Simple points revisited

Suppose that A is an integral domain with field of fractions F . Let
R = (ri,j ) be an m × n matrix with entries in A. Consider the A-module
n
M Xn
MA (R) := Aei /h ri,j ej | i = 1, . . . , mi
j=1 j=1
given by generators and relations. If Y is an invertible m × m matrix

with entries in A, then the change of basis argument gives MA (Y R) ∼ =
MA (R). Similarly, if Z is an invertible n × n matrix with entries in A,
then MA (RZ) ∼ = MA (R). Now by linear algebra we can find invertible
matrices Y and Z with entries in F such that

Ir 0
R=Y Z,
0 0
84 Tangent spaces
where r is the rank of R. Putting all entries of Y and Z over a common
denominator, we may assume that Y and Z have entries in Af for some
non-zero f ∈ A. Note that
MA (R)f ∼
= MAf (R).
So MA (R)f is a free Af -module of rank n − r.

Recall that if X is an affine variety, we write ΩX for Ωk[X]/k . If x ∈ X,
let us also write ΩX (x) for the vector space kx ⊗k[X] ΩX . This is called
cotangent space for ΩX (x) ∼ = (Tx X)∗ . Indeed, using Example 6.5.3, we
have
Tx X = Homk[X] (ΩX , kx ) ∼
= Homk (kx ⊗k[X] ΩX , k) = ΩX (x)∗ .
∂f
If k[X] = k[T1 , . . . , Tn ]/(f1 , . . . , fm ), let R be the m×n matrix ( ∂Tji (t))
∂f
and R(x) = ( ∂Tji (x)). Then ΩX = Mk[X] (R) and ΩX (x) = Mk (R(x)).
Lemma 6.6.1 Assume that X is an irreducible affine variety.

(i) dimk(X) Mk(X) (R) = dim X.
(ii) If x ∈ X is a simple point, then there is f ∈ k[X] with f (x) 6= 0
such that Mk[X] (R)f is a free k[X]f -module of rank dim X with
basis given by dim X out of the images of the ei .
Proof (i) Since k is algebraically closed, k(X) is a separable extension

of k. So Theorem 6.5.6 tells us that dim X = dimk(X) Ωk(X)/k . But
Ωk(X)/k = k(X) ⊗k[X] ΩX ∼ = Mk(X) (R).
(ii) In view of (i), the rank of the matrix R is r := n − dim X. Some
r×r-minor of R(x) has non-zero determinant. Reordering if necessary we
may assume that this is the principal minor in the top left hand corner.
Let f be the determinant of this minor, so f (x) 6= 0. On localizing at
f , the matrix R becomes equivalent to

Ir 0
.
0 0
The lemma implies
Theorem 6.6.2 Let X be an irreducible variety. If x ∈ X is a simple

point, there is an affine neighborhood U of x such that ΩU is a free
k[U ]-module on basis dg1 , . . . , dgdim X for suitable gi ∈ k[U ].
6.7 Separable morphisms 85
6.7 Separable morphisms
Recall that if X is an affine variety, we write ΩX for Ωk[X]/k and ΩX (x)
for the cotangent space kx ⊗k[X] ΩX at X.
Let ϕ : X → Y be a separable dominant morphism of irreducible affine
varieties. The composition of ϕ∗ : k[Y ] → k[X] and dX : k[X] → ΩX is
a derivation
dX ◦ ϕ∗ : k[Y ] → ΩX .
So by the universal property of the differentials we get induced a k[Y ]-

module map
ϕ̂∗ : ΩY → ΩX
such that dX ◦ ϕ∗ = ϕ̂∗ ◦ dY .

Let x ∈ X and y = ϕ(x). The k[X]-module kx viewed as a k[Y ]-
module via ϕ∗ is ky . After identifying Tx X with Homk[X] (ΩX , kx ) and
Ty Y with Homk[Y ] (ΩY , ky ), the map dϕx becomes:
dϕx : Homk[X] (ΩX , kx ) → Homk[Y ] (ΩY , ky ), θ 7→ θ ◦ ϕ̂∗ .
Theorem 6.7.1 Let ϕ : X → Y be a morphism of irreducible varieties.

(i) Assume that x ∈ X and y = ϕ(x) ∈ Y are simple points and that
dϕx is surjective. Then ϕ is a dominant separable morphism.
(ii) Assume that ϕ is a dominant separable morphism. Then the
simple points x ∈ X with ϕ(x) simple and dϕx surjective form a
non-empty open subset of X.
Proof (i) We may assume that X and Y are affine and ΩX , ΩY are
free K[X]-, resp. k[Y ]-modules of rank d = dim X resp. e = dim Y .
In particular X and Y are smooth. The map ϕ̂∗ : ΩY → ΩX of k[Y ]-
modules induces a homomorphism of free k[X]-modules
ψ : k[X] ⊗k[Y ] ΩY → ΩX .
We can represent ψ as a d × e-matrix A with entries in k[X], fixing bases

for ΩX and ΩY . Suppose that dϕx is surjective. Then A(x), which
represents the dual map dϕ∗x : ΩY (y) → ΩX (x), is injective, hence a
matrix of rank e. Hence the rank of A itself is at least e, hence equal to
e since rank cannot be more than the number of columns. This shows
that ψ is injective. Hence ϕ̂∗ is injective too. Since ΩX and ΩY are free
modules, this implies that ϕ∗ : k[Y ] → k[X] must be injective. So ϕ
must be dominant.
86 Tangent spaces
Moreover, injectivity of ψ implies the injectivity of
k(X) ⊗k[Y ] ΩY → k(X) ⊗k[X] ΩX .
This is the map α in the exact sequence

α β
k(X) ⊗k(Y ) Ωk(Y )/k −→ Ωk(X)/k −→ Ωk(X)/k(Y ) −→ 0.
Hence k(X) is a separable extension of k(Y ) by the differential criterion

for separability.
Example 6.7.2 I will illustarte the usefulness of the theorem by an

example from my research. Recently Jon Brundan and I needed to
establish the following.
Consider the polynomial algebra
[r]
C[xij | 1 ≤ i, j ≤ n, r = 1, . . . , l],
and let
(r) [s ] [s ] [s ]
X X
yi,j = xi01,i1 xi12,i2 · · · xir−1
r
,ir
1≤s1 <···<sr ≤l 1≤i0 ,··· ,ir ≤n
i0 =i,ir =j
In order to complete the proof of a theorem, we needed to show that the

(r)
elements {yi,j }1≤i,j≤n,r=1,...,l are algebraically independent.
[r]
Let us identify C[xij ] with the coordinate algebra C[Mn×l ] of the affine
[r]
variety Mn×l of l-tuples (A1 , . . . , Al ) of n × n matrices, so that xi,j is the
function picking out the ij-entry of the rth matrix Ar . Let θ : Mn×l →
Mn×l be the morphism defined by (A1 , . . . , Al ) 7→ (B1 , . . . , Bl ), where
Br is the rth elementary symmetric function
X
er (A1 , . . . , Al ) := As1 · · · Asr
1≤s1 <···<sr ≤l
[r] (r)
in the matrices A1 , . . . , Al . The comorphism θ∗ maps xi,j to yi,j . So to
(r)
show that the yi,j are algebraically independent, we need to show that
θ∗ is injective, i.e. that θ is a dominant morphism of affine varieties.
For this it suffices to show that the differential of θ is surjective at some
point x ∈ Mn×l .
Pick pairwise distinct scalars c1 , . . . , cl ∈ C and consider
x := (c1 In , . . . , cl In ).
6.8 Problems 87
Identifying the tangent space Tx (Mn×l ) with the vector space Mn⊕l , a cal-
culation shows that the differential dθx maps (A1 , . . . , Al ) to (B1 , . . . , Bl )
where
Xl
Br = er−1 (c1 , . . . , cbs , . . . , cl )As .
s=1
Here er−1 (c1 , . . . , cbs , . . . , cl ) denotes the (r − 1)th elementary symmetric

function in the scalars c1 , . . . , cl excluding cs . We just need to show this
linear map is surjective, for which it clearly suffices to consider the case
n = 1. But in that case its determinant is the Vandermonde determinant
Q
1≤r<s≤l (cs −cr ), so it is non-zero by the choice of the scalars c1 , . . . , cl .
6.8 Problems
Problem 6.8.1 Let X ⊂ An be a closed subset, I = I(X), and J be

the ideal of k[T1 , . . . , Tn ] generated by all dx f for f ∈ I. If f1 , . . . , fl
generate I, then dx f1 , . . . , dx fl generate J.
7
Complete Varieties
7.1 Main Properties

A variety X is called complete if for any variety Y the projection π2 :
X × Y → Y is a closed map.
Remark 7.1.1 Completeness is an algebraic analogue of compactness.

To be more precise, let X be a locally compact Hausdorff topological
space. One can prove that X is compact if and only if for any locally
compact space Y the projection π2 : X × Y → Y is closed.
Example 7.1.2
(i) A point is complete, as if X is a point, π2 : X × Y → Y is an
isomorphism.
(ii) A1 is not complete. Indeed, take Z = Z(T1 T2 − 1) ⊂ A1 × A1 =
A2 . Then π2 maps Z onto A1 \ {0}.
Remark 7.1.3
(i) X is complete if and only if all its irreducible components are
complete.
(ii) X is complete if for any irreducible affine variety Y the projection
π2 : X × Y → Y is closed.
Proposition 7.1.4 Let X, Y be varieties.

(i) If X is complete and Y ⊂ X is closed then Y is complete.
(ii) If X and Y are complete, then so is X × Y .
(iii) If ϕ : X → Y is a morphism and X is complete, then ϕ(X) is
closed and complete.
(iv) If Y is a complete subvariety of X, then Y is closed.
88
7.2 Completeness of projective varieties 89
(v) If X is complete and irreducible, then OX (X) = k. In particular,
if X is complete and affine, then X is a finite number of points.
Proof (i) A closed subset of Y × Z is also closed in X × Z.

(ii) Projection X × Y × Z is a composition of πY × idZ : X × Y × Z →
Y × Z and πZ : Y × Z → Z.
(iii) Since Y is a variety, the graph of ϕ is closed in X × Y . Its image
is ϕ(X), which is closed by completeness of X. To show completeness
of ϕ(X), take a closed subset K ⊂ ϕ(X) × Z for some Z. Consider
projections π2 : X × Z → Z, π20 : ϕ(X) × Z → Z, and note that
π20 (K) = π2 ((ϕ × idZ )−1 (K)).
(iv) Apply (iii) to the embedding of Y into X.
(v) Let f ∈ OX (X). Then f is a morphism from X to A1 , cf. Prob-
lem 4.6.2. By (iii), f (X) is closed complete irreducible subvariety of A1 ,
and it could not be A1 itself, since A1 is not complete, so f (X) is a
point, i.e. f is a constant.
7.2 Completeness of projective varieties
Theorem 7.2.1 Any projective variety is complete.
Proof In view of Proposition 7.1.4(i) and Remark 7.1.3(ii), it suffices to

prove that π2 : Pn × Y → Y is closed for any irreducible affine variety
Y . Set R := k[Y ].
For 0 ≤ i ≤ n, let Pni be the affine open set of Pn given by Xi 6= 0,
where X0 , X1 , . . . , Xn are the coordinate ‘functions’ on Pn . Then the
affine open sets Ui := Pni × Y cover Pn × Y . Moreover, we can identify
k[Ui ] with Ri := k[X0 /Xi , . . . , Xn /Xi ] ⊗ R = R[X0 /Xi , . . . , Xn /Xi ].
Let Z be any cosed set in Pn × Y , and y ∈ Y \ π2 (Z). We want to find
a neighborhood of y in Y of the form Yf which is disjoint from π2 (Z).
This amounts to finding f ∈ R with f 6∈ M := My and such that f
vanishes on π2 (Z). Let π2i := π2 |Ui and Zi := Z ∩ Ui , 0 ≤ i ≤ n. Now
f |π2 (Z) ≡ 0 is equivalent to the statement that the pullback of (π2i )∗ (f )
is zero on Zi ⇔ (π2i )∗ (f ) ∈ I(Zi ) / Ri . The existence of such f will
follow from Nakayama’s Lemma applied to a suitable R-module, which
we now construct.
First consider the polynomial ring S := R[X0 , . . . , Xn ] with natural
grading S = ⊕m Sm . We construct the homogeneous ideal I /S by letting
90 Complete Varieties
Im consist of all f (X0 , . . . , Xn ) ∈ Sm such that f (X0 /Xi , . . . , Xn /Xi ) ∈
I(Zi ) for each i.
Next, fix i and let f ∈ I(Zi ). We claim that the multiplication of f
by a sufficiently high power of Xi will take f into I. Indeed, for large
m, Xim f becomes a homogeneous polynomial of degree m. Moreover,
(Xim /Xjm )f ∈ Rj vanishes on Zi ∩ Uj = Zj ∩ Ui , while (Xim+1 /Xjm+1 )f
vanishes at all points of Zj not in Ui . So (Xim+1 /Xjm+1 )f vanishes on
Zj . Since j is arbitrary, we conclude that Xim+1 f lies in Im+1 .
Now, Zi and Pni × {y} are disjoint closed subsets of the affine variety
Ui , so their ideals I(Zi ) and M Ri generate Ri , i.e. we can write 1 =
P
fi + j mij gij , where fi ∈ I(Zi ), mij ∈ M , and gij ∈ Ri . By the
preceding paragraph, multiplication by a sufficiently high power of Xi
takes fi into I. We can choose this power large enough to work in
these equations for all fi and to take gij into S as well. So we obtain
Xim ∈ Im + M Sm for all i. Enlarging m even more, we can get all
monomials of degree m in X0 , . . . , Xn to lie in Im + M Sm . This implies
Sm = Im + M Sm .
Now apply Corollary 2.1.9 to the finitely generated R-module Sm /Im ,
which satisfies M (Sm /Im ) = Sm /Im . The conclusion is that there exists
f ∈ R \ M such that f annihilates Sm /Im , i.e. f Sm ⊂ Im . In particular,
f Xim ∈ Im , so by definition of Im we have (f Xim )(X0 /Xi , . . . , Xn /Xi ) ∈
I(Zi ), but (f Xim )(X0 /Xi , . . . , Xn /Xi ) ∈ I(Zi ) = f .
Part two
Algebraic Groups
8
Basic Concepts
8.1 Definition and first examples
Definition 8.1.1 An algebraic group is an affine variety G equipped

with morphisms of varieties µ : G × G → G, ι : G → G that give G the
structure of a group. A morphism f : G → H of algebraic groups is a
morphism of varieties that is a group homomorphism too.
It is possible to consider algebraic groups which are not necessarily

affine varieties, so strictly speaking one we should have used the term
affine algebraic group above. As we will only meet affine algebraic groups
we will drop the word ‘affine’.
The kernel of a morphism f : G → H of algebraic groups is a closed
subgroup of G, so it is an algebraic group in its own right. The same
will turn out to be true about the images.
Translation by an element g ∈ G is an isomorphism of varieties, so all
geometric properties at one point can be transferred to any other point.
For example, as G has simple points, G is smooth.
Example 8.1.2 (i) The additive group Ga is the group (k, +), i.e. the
affine variety A1 under addition.
(ii) The multiplicative group Gm is the group (k × , ×), i.e. the principal
open subset A1 \ {0} under multiplication.
(iii) The group GLn = GLn (k) is the group of all invertible n × n ma-
2
trices over k. As a variety, this is a principal open set in Mn (k) = An
corresponding to the determinant. Since the formulas for matrix mul-
tiplication and inversion involve only polynomials in the matrix entries
and 1/ det, the group structure maps are morphisms of varieties.
Let V be an n-dimensional vector space over k. Then by fixing a
93
94 Basic Concepts
basis we can define a structure of an algebraic group on GL(V ) which
is independent of the choice of basis. Of course, GL(V ) ∼= GLn .
(iv) The group SLn = SLn (k) is the closed subgroup of GLn defined
by the zeros of det −1.
(v) The group Dn of invertible diagonal matrices is a closed subgroup
of GLn (given be the zeros of which functions?) It is isomorphic to the
direct product Gm × · · · × Gm (m copies).
(vi) The group Un of upper unitriangular matrices is another closed
subgroup of GLn .
(vii) The orthogonal group On = {x ∈ GLn | xxt = 1}. We exclude
the characteristic 2 when considering this example...
(viii) The special orthogonal group SOn = On ∩ SLn is a normal
subgroup of On of index 2.
(ix) The symplectic group Sp2n = {x ∈ GLn | xt Jx = J} where

0 In
−In 0
is another closed subgroup.
Let G be an (affine) algebraic group with the identity element e, and

put A = k[G]. The map
ε : A → k, f 7→ f (e)
is an algebra homomorphism (called augmentation). Consider also the

dual morphisms
∆ := µ∗ : A → A ⊗ A
(called comultiplication) and
σ := ι∗ : A → A
(called antipode). In follows using group axioms that these define the
structure of a Hopf algebra on k[G]. Conversely, a structure of the Hopf
algebra on k[G] defines a structure of an algebraic group on G. An easy
corollary of Theorem 3.5.1 now is that the categories of (affine) algebraic
groups and affine Hopf algebras are contravariantly equivalent.
Example 8.1.3 (i) k[Ga ] = k[T ] with ε(T ) = 0, σ(T ) = −T , and

∆(T ) = T ⊗ 1 + 1 ⊗ T .
(ii) k[Gm ] = k[T, T −1 ] with ε(T ) = 1, σ(T ) = T −1 , and ∆(T ) = T ⊗T .
(iii) k[GLn ] = k[Ti,j | 1 ≤ i, j ≤ n]det with ε(Tij ) = δij , σ(Tij ) =
(−1)i+j Mj,i / det (where Mj,i is the determinant of the (j, i)minor), and
Pn
∆(Ti,j ) = k=1 Tik ⊗ Tkj .
A rational representation of G in a finite dimensional k-vector space

V is a homomorphism of algebraic groups ρ : G → GL(V ). The notion
of a rational representation is equivalent to that of a rational G-module:
V is called a rational G-module if it is a G-module in the usual sense and
the corresponding representation is rational. From the point of view of
Hopf algebras the notion of a G-module is equivalent to the notion of a
comodule over the Hopf algebra k[G] (read about this notion somewhere
or better yet invent it yourself!)
8.2 First properties

Let G be an algebraic group. We note that only one irreducible compo-
nent of G can pass through the identity element e. Indeed, if X1 , . . . , Xm
are the distinct irreducible components of G containing e. The image
of the irreducible variety X1 × · · · × Xm under the product morphism
is an irreducible subset X1 . . . Xm of G, which again contains e. So
X1 . . . Xm lies in some Xi . On the other hand each of the components
X1 , . . . , Xm clearly lies in X1 . . . Xm . This forces m = 1. Denote by
G◦ this unique irreducible component of G containing e, and call it the
identity component of G.
Proposition 8.2.1 Let G be an algebraic group.

(i) G◦ is a normal subgroup of finite index in G, whose cosets are
the connected as well as irreducible components of G.
(ii) Each closed subgroup of finite index in G contain G◦ .
Proof (i) We have ι(G◦ ) is an irreducible component of G containing e,

so ι(G◦ ) = G◦ . It also follows from the argument preceding the theorem
that G◦ G◦ = G◦ , so G◦ is a (closed) subgroup of G.
For any x ∈ G, xG◦ x−1 is also an irreducible component of G con-
taining e, so xG◦ x−1 = G◦ , i.e. G◦ is normal. Its cosets are translates
of G◦ , hence must also be irreducible components of G. As there are
only finitely many irreducible components, it follows that [G : G◦ ] < ∞.
Since the cosets are disjoint, they are also connected components of G.
(ii) If H is a closed subgroup of a finite index in G, then H ◦ is a
closed subgroup of finite index in G◦ , and each of its finitely many left
96 Basic Concepts
cosets in G◦ is also closed, and so the union of the cosets distinct from
H ◦ is closed. Hence H ◦ is also open in G◦ . Since G◦ is irreducible it is
connected, whence H ◦ = G◦ .
The algebraic group is called connected if G◦ = G.
Lemma 8.2.2 Let U and V be dense open subsets of G. Then G = U V .
Proof Let x ∈ G. Then xV −1 and U are dense open subsets. So they

have to meet, forcing x ∈ U V .
Lemma 8.2.3 Let H < G be a subgroup of an algebraic group G. Then:

(i) H̄ is a subgroup of G.
(ii) If H is constructible, then H = H̄.
(iii) If H contains a dense open subset of H̄, then H = H̄.
Proof (i) As ι is a homeomorphism, we have ι(H̄) = ι(H) = H̄. Simi-

larly, translation by x is a homeomorphism, so xH = xH̄, i.e. H H̄ ⊂ H̄.
Therefore, if x ∈ H̄, we have Hx ⊂ H̄, so H̄x = Hx ⊂ H̄.
(ii),(iii) If H is constructible, it contains a dense open subset U of
H̄, see Problem 5.5.3. Then H is also open in H̄, as H is a union of
translates of U . By Lemma 8.2.2, H̄ = HH = H.
Corollary 8.2.4 Let A, B be closed subgroups of G. If B normalizes A,

then AB is a closed subgroup of G.
Proof It is clear that AB is a subgroup. Moreover, it the image of A×B

under the product morphism, hence constructible by Theorem 5.3.3,
hence closed by the lemma.
Lemma 8.2.5 Let ϕ : G → H be a morphism of algebraic groups. Then:

(i) ker ϕ is a closed subgroup of G.
(ii) im ϕ is a closed subgroup of H.
(iii) ϕ(G◦ ) = ϕ(G)◦ .
(iv) dim G = dim ker ϕ + dim im ϕ.
Proof (i) follows from the continuity of ϕ and (ii) follows from The-
orem 5.3.3 and Lemma 8.2.3(ii). Now, ϕ(G◦ ) is closed by (ii) and ir-
reducible, hence lies in ϕ(G)◦ . Being of finite index in ϕ(G), it must
equal ϕ(G)◦ , thanks to Proposition 8.2.1(ii). Finally, Theorem 5.3.1(ii)
implies that dim G−dim ϕ(G) = dim ϕ−1 (x) for ‘most’ points x ∈ ϕ(G).
But all fibers ϕ−1 (x) are isomorphic to ker ϕ, so we have (iv).
Proposition 8.2.6 Let (Xi , ϕi )i∈I be a family of irreducible varieties

and morphisms ϕi : Xi → G such that e ∈ Yi := ϕi (Xi ) for all i. Let H
be the smallest subgroup of G containing all Yi . Then:
(i) H is closed and connected.
(ii) H = Yaε11 . . . Yaεnn for some a1 , . . . , an ∈ I and ε1 , . . . , εn ∈ {±1}.
Proof We may assume that the sets Yi−1 occur among the Yj . Note that
for each a := (a1 , . . . , an ) ∈ I n , Ya := Ya1 . . . Yan is irreducible, hence Ȳa
is irreducible, too. Obviously Ya Yb = Y(a,b) .
Moreover, Ȳa Ȳb ⊂ Y(a,b) . Indeed, for x ∈ Yb , the homeomorphism (of
G) y 7→ yx sends Yb to Y(a,b) , hence Ȳb into Y(a,b) , i.e Ȳa Yb ⊂ Y(a,b) . Now
x ∈ Ȳa sends Yb into Y(a,b) , hence Ȳb as well.
Now choose the tuple a such that dim Ya is maximal. As e ∈ Ya , we
have for any b that Ȳa ⊂ Ȳa Ȳb ⊂ Y(a,b) . Equality holds by dimensions,
so Ȳb ⊂ Ȳa for every b, and Ȳa is closed under multiplication. Choosing
b such that Yb = Ya−1 , we also have Ȳa stable under inversion. So Ȳa is a
group. Since Ya is constructible, it contains a dense open subset of Ȳa ,
whence Ȳa = Ya Ya in view of Lemma 8.2.2.
Finally, we claim that H = Ȳa . It is clear that H is contained in Ȳa ,
as we know that each Yb ⊂ Ȳa . Since H ⊃ Ya , we have H̄ = Ȳa . Finally,
H ⊃ Ya also implies that H contains a dense open subset of H̄, so H is
closed by Lemma 8.2.3(iii).
Corollary 8.2.7 Assume that (Gi )i∈I is a family of closed connected

subgroups of G. Then the group H generated by them is closed and
connected. Furthermore, H = Ga1 . . . Gan for some a1 , . . . , an ∈ I.
Example 8.2.8 It is easy to see that the groups Gm , Ga , GLn are con-
nected. It is less obvious that SLn , Sp2n , and SOn are connected. This
can be deduced using Corollary 8.2.7 and some group theory. For ex-
ample the group SLn is known to be generated by transvections. It
follows that the subgroups Gij = {E + tEij | t ∈ k} generate SLn .
This transvection subgroups are closed and isomorphic to Ga , hence
connected. For Sp2n , let V be the 2n-dimensional vector space on
which Sp2n acts, and (·, ·) be the non-degenerate symplectic bilinear
form preserved by the group. For v ∈ V \ {0} define the symplectic
transvection group Gv to consist of all linear transformations of the form
98 Basic Concepts
w 7→ w + t(w, v)v (t ∈ k). It remains to use the known fact that the Gv
generate Sp2n . A similar proof is available for SOn .
As SOn is of index 2 in On , it follows that it is the identity component
of On .
Corollary 8.2.9 Let H and K be closed subgroups of G with H conncted.

Then the commutator group (H, K) generated by all commutators [h, k]
with h ∈ H, k ∈ K , is closed and connected.
Proof Take the index set I in the proposition to be K and the maps
ϕk : H → G to be the maps h 7→ hkh−1 k −1 (k ∈ K).
Example 8.2.10 Recall the definition of the derived series
G = G(0) ≥ G(1) ≥ . . .
of a group G: G(0) = G, G(i+1) = (G(i) , G(i) ). The group G is the called

solvable if G(i) = {e} for some i. In case G is a connected algebraic
group, each of the derived subgroups are closed connected subgroup of
G. So either G(i+1) = G(i) or dim G(i+1) < dim G(i) . Thus we see that
for algebraic groups the derived series stabilizes after finitely many steps.
Similar remarks apply to nilpotent algebraic groups.
8.3 Actions of Algebraic Groups

Let G be an algebraic group and X be a variety (not necessarily affine).
We say that G acts on X, or that X is a G-variety, if we are given a
morphism
G × X → X, (g, x) 7→ gx
of varieties that makes X into a G-set in the usual sense. If the G-action
on X is transitive, X is called a homogeneous space.
Lemma 8.3.1 Let G act on X. Let Y, Z be subsets of X with Z closed.

(i) The set {g ∈ G | gY ⊂ Z} is closed; in particular NG (Z) := {g ∈
G | gZ ⊂ Z} is closed.
(ii) For each x ∈ X the stabilizer Gx is a closed subgroup of G; in
particular, CG (Y ) := {g ∈ G | gy = y for any y ∈ Y } is closed.
(iii) The fixed point set X g of g ∈ G is closed in X; in particular X G
is closed.
8.3 Actions of Algebraic Groups 99
Proof (i) For each y ∈ X the orbit map fy : G → X, g 7→ gy is a
morphism. So fy−1 (Z) is closed in G. Now note that
{g ∈ G | gY ⊂ Z} = ∩y∈Y fy−1 (Z).
(ii) Observe that Gx = {g ∈ G | g{x} ⊂ {x}} and apply (i).

(iii) Consider the morphism ψ : X → X × X, x 7→ (x, gx). Then X g
is the inverse image under ψ of the diagonal, which is closed, since X is
a variety.
Remark 8.3.2 The lemma shows that things like centralizers of sub-
sets, normalizers of closed subsets, fixed point sets , etc. are closed.
However orbits themselves are not closed in general. In fact the struc-
ture of orbits of an algebraic group on a variety can be very interesting.
Also, connectedness of centralizers and normalizers is not to be taken
for granted.
Theorem 8.3.3 Let G act on X. Then each orbit is smooth, locally

closed subset subset of X, whose boundary Gx − Gx is a union of orbits
of strictly smaller dimension. In particular, orbits of minimal dimension
are closed (so closed orbits exist). If G is connected, the orbits are
irreducible.
Proof Let O = Gx. As the image of G under the orbit map, O is

constructible, hence contains an open dense subset U of Ō. (Also, O
is irreducible if G is connected.) But G acts transitively on O (leaving
Ō stable), so O = ∪g∈G gU is open in Ō, and O is smooth. Therefore
Ō − O is closed and of strictly lower dimension than dim Ō = dim O.
Being G-stable, this boundary is the union of other G-orbits.
Example 8.3.4 Let G = GLn = GL(V ) where V = k n (viewed as an

affine n-space). There are just two orbits of G on V : the ponit {0} and
the rest V − {0}, an open orbit of dimension n. What can you say about
stabilizers in this action? More generally, if V is a rational G-module
over an arbitrary algebraic group G, then v 7→ gv defines a structure of
G-variety on V ∼ = Adim V .
Example 8.3.5 Again take G = GLn = GL(V ) and define the G-action
on P(V ) via ghvi = hgvi (here hvi denote the line spanned by a non-zero
vector v ∈ V ). In other words this is just the natural action of GLn on
100 Basic Concepts
the lines of V , which is transitive by linear algebra. What can you say
about stabilizers in this action?
In order to check that this is an action in the sense of algebraic groups,
we need to check that the corresponding map ρ : G × P(V ) → P(V )
is a morphism of varieties. For this we employ the Affine Criterion
(Theorem 4.2.4) with the usual affine open subsets Vi of P(V ) (0 ≤ i ≤
n), and Ui = ϕ−1 (Vi ) (Here ϕ : G × Pn → Pn is the action map.)
Let V 0 = V − {0} be the non-trivial G-orbit from Example 8.3.4. it is
easy to check using Affine Criterion that the map V 0 → P(V ), v 7→ hvi
is a G-equivariant morphism of varieties.
Example 8.3.6 The natural actions of G = GLn = GL(V ) on the

Grassmann variety Gd (V ) and the flag variety F are transitive by linear
algebra. These actions are morphic as they are just restrictions of the
action of G on P(Λd (V )) and P(Λ1 (V )) × · · · × P(Λn−1 (V )) × P(Λn (V )),
respectively. What can you say about stabilizers in this actions?
Lemma 8.3.7 Let G be a connected algebraic group and X, Y be homoge-

neous spaces over G. Suppose ϕ : X → Y is a G-equivariant morphism.
Set r = dim X − dim Y . Then:
(i) ϕ is surjective and open.

(ii) for each closed irreducible subset W ⊂ Y all irreducible compo-
nents of ϕ−1 (W ) have dimension r + dim W .
Proof Surjectivity is clear. Now, it follows from (ii) and Theorem 5.4.1
that ϕ is open. It remains to prove (ii). By Theorem 5.3.1, there is
an open set U ⊂ Y such that for each irreducible closed subset W ⊂ Y
meeting U , the components of ϕ−1 (W ) meeting ϕ−1 (U ) have dimensions
dim W + r. Since G acts transitively on Y and X, the G-translates of U
cover Y and the G-translates of ϕ−1 (U ) cover X. This implies (ii).
8.4 Linear Algebraic Groups

A linear algeraic group is a closed subgroup of some GLn . The following
theorem can be thought of as the analogue of the famous theorem that
any finite group is a subgroup of some symmetric group Sn .
Theorem 8.4.1 Every (affine) algebraic group is linear.

8.4 Linear Algebraic Groups 101
To prove the theorem we need to find a finite dimensional vector space
on which G acts, and the only place we can look for it is inside the regular
module k[G]. Given g ∈ G, the map G → G, h 7→ hg is a morphism of
varieties, whose dual map is ρg : k[G] → k[G], where
ρg (f )(h) = f (hg) (f ∈ k[G], h ∈ G).
This defines a representation ρ of G in the (usually infinite dimensional
space) k[G], called (right) regular representation or representation by
right translations of functions. The left regular representation λ is de-
fined similarly via
λg (f )(h) = f (g −1 h) (f ∈ k[G], h ∈ G).
The antipode map is actually an isomorphism of the left and right regular
representations, so we will usually refer to it as the regular representation
and use the right one if we need to write some formulas. The following
lemma will help us to deal with the problem of infinite dimensionality
of k[G].
Lemma 8.4.2 The regular representation is locally finite dimensional,

i.e. every element of k[G] is contained in a finite dimensional submodule.
Proof Let us take a non-zero f ∈ k[G]. Let W be the subspace of k[G]

spanned by all right translations ρg f . We need to show that W is finite
Pn
dimensional. Write ∆f = i=1 fi ⊗ gi . Let X be the finite dimensional
ubspace of k[G] spanned by all fi . Now consider x ∈ G. We have
n
X
(ρx f )(h) = f (hx) = (∆f )(h, x) = fi (h)gi (x).
i=1
Pn
Hence ρx f = i=1 gi (x)fi ∈ X. Hence W ⊂ X and W is finite dimen-
sional.
Proof of the theorem Choose linearly independent generators f1 , . . . , fn

of the algebra k[G]. Applying the lemma, we may assume (adding finitely
many more generators if necessary) that the span E of the fi is invariant
under all right translations. Now consider the restriction
ψ : G → GL(E), x 7→ ρx |E
of ρ.
P
Fix i and write ∆fi = j gj ⊗ hj with gj linearly independent and
P
hj 6= 0. As in the proof of the lemma, ρx fi = hj (x)gj for all x ∈ G,
102 Basic Concepts
which implies gj ∈ E, so we can write
X
∆fi = fj ⊗ hij (1 ≤ i ≤ n). (8.1)
j
Then the coordinates of the matrix of ψ(x) with respect to the basis
f1 , . . . , fn are hij (x). Hence ψ is a morphism of varieties.
P
Next notice that fi (x) = fi (ex) = j fj (e)hi,j (x), so
X
fi = fj (e)hi,j . (8.2)
j
If ψ(x) = e, then hi,j (x) = δi,j , so fi (x) = fi (e) for all i, whence x = e,
as fi ’s generate k[G].
By Lemma 8.2.5(ii), G0 := im ψ is a closed subgroup of GL(E). To
complete the proof, we need only to show that ψ : G → G0 is an isomor-
phism of varieties, i.e. ψ ∗ : k[G0 ] → k[G] is an isomorphism of algebras.
As ψ is surjective, ψ ∗ is injective. On the other hand, let tij be coordi-
nate functions on GL(E) restricted to G0 . Note that ψ ∗ (tij ) = hij , and
the hij generate k[G] in view of (8.2), so ψ ∗ is surjective.
8.5 Problems
Problem 8.5.1 Let A be a finite dimensional k-algebra. Show that

Aut(A) is a closed subgroup of GL(A).
Solution. Aut(A) is the stabilizer of an element t ∈ A∗ ⊗ A∗ ⊗ A, see

the proof of Corollary 9.5.2.
Problem 8.5.2 Describe Aut(Gm ), Aut(Ga ), and End (Gm , Gm ).
Solution. Working with k[G], we get Aut(Gm ) ∼ = Z2 , where the only non-
trivial automorphism is z 7→ z −1 . Moreover, End (Gm ) ∼
= Z with m ∈ Z
corresponding to the endomorphism z 7→ z . Finally, Aut(Ga ) ∼
m
= k× ,
×
with a ∈ k corresponding to the endomorphism z 7→ az.
Problem 8.5.3 Closed subset of G containing e and closed under mul-

tiplication is a subgroup of G.
Solution. Let X be the subset and x ∈ X. Consider the morphism

ϕ : X → X, y 7→ yx. It suffices to show that this morphism is surjective,
as then e is in the image, and the result follows.
8.5 Problems 103
In order to prove that ϕ is surjective, let Z be an irreducible compo-
nent of X of maximal dimension. Then ϕ(X) is irreducible of the same
dimension, as ϕ is the restriction to X of an autmorphism of G. So ϕ(Z)
must be an irreducible component of X. This proves that ϕ permutes
irreducible components of X. As X is one-to-one, this argument can
now be applied again to the irreducible components of the next largest
dimension, etc.
Problem 8.5.4 Let N < GLn be the group of monomial matrices, i.e.
matrices having precisely one non-zero entry in each column and each
entry. Prove that N ◦ is the subgroup of all diagonal matrices in GLn .
Solution. Humphreys, problem 7 after section 7. The group D of diag-
onal matrices is connected, and [N : D] is finite.

1 0
Problem 8.5.5 Show that the subgroup of GL2 (C) geberated by
0 −1

1 1
and is not closed.
0 −1
Solution. Let X =∼ A1 ⊂ GL2 (C) be the closed subset which consists of
all upper unitrangular matrices. Note that our subgroup intersects X
at the subset of all upper-unitrangular matrices with integer entries in
the corner. This is not closed, as Z ⊂ A1 is not closed.
Problem 8.5.6 Let G be a connected algebraic group. Prove that any

finite normal subgroup H lies in the center of G.
Solution. If h ∈ H, then the image of the morphism G → G, x 7→ xhx−1
is connected and contained in H, so the image is trivial.
Problem 8.5.7 True or false? Let ϕ : G → H be a morphism of

algebraic groups which is an isomorphism of abstract groups. Then ϕ is
an isomorphism of algebraic groups.
Solution. False: consider Fr : Gm → Gm or Problem 8.5.8(iii).
Problem 8.5.8 We have A := k[SL2 ] = k[T11 , T12 , T21 , T22 ]/(T11 T22 −
T12 T21 − 1) = k[t11 , t12 , t21 , t22 ] (tij denoting the image of Tij ). Let B
be the subalgebra of A generated by all products tij tkl .
(i) Show that B be is a Hopf subalgebra of A and deduce that that
there is an algebraic group P SL2 whose algebra is B. Show that
104 Basic Concepts
the inclusion map B → A defines a surjective homomorphism of
algebraic groups SL2 → P SL2 with kernel of order at most 2.
(ii) If char k 6= 2, then B is the algebra of functions f ∈ A such that
f (−X) = f (X) for all X ∈ SL2 .
(iii) If char k = 2, then the homomorphism of (i) defines an isomor-
phism of underlying abstract groups but is not an isomorphism
of algebraic groups.
Solution. (i) That B is a Hopf subalgebra is easily checked using ex-
plicit formulas for coproduct and antipode. Also B is a reduced finitely
generated k-algebra, so it corresponds to an algebraic group by general
principles. Also, the inclusion map ι : B → A, being a Hopf algebra
map, defines a surjective homomorphism ι∗ : SL2 → P SL2 .
Now B is generated by the elements
t211 , t11 t12 , t11 t21 , t11 t22 , t212 , t12 t22 , t221 , t21 t22 , t222 ,
as t11 t22 = t12 t21 + 1. Now, using the counit, we see that the identity e
in P SL2 is defined by equations t211 (e) = 1, t222 (e) = 1, t11 t22 (e) = 1 and
tij tkl (e) = 0 for all other generators. So A = (aij ) maps to e under ι∗ if
and only if a211 = 1, a222 = 1, a11 a22 = 1 and aij akl = 0 for all other pairs
of indices corresponding to the generators. It follows that the kernel of
ι∗ is ±I.
(ii) Direct check.
(iii) If char k = 2, ι∗ is bijective. Of course it is not an isomorphism
since ι is not surjective.
Problem 8.5.9 Let X be a G-variety and a : G × X → X is the action

map. Define the left action of G on k[X] via
(gf )(x) = f (g −1 x) (g ∈ G, x ∈ X, f ∈ k[X]).
Note that this yields a representation of abstract group G in k[X].
(i) The representation is locally finite dimensional.
(ii) A finite dimensional subspace V ⊂ k[X] is G-stable if and only if
a∗ (V ) ⊂ k[G] ⊗ V . If so, the action of G on V defines a rational
representation of G.
(iii) There is a sequence of finite dimensional G-submodules Vi ⊂ k[G]
such that V1 ⊂ V2 ⊂ . . . and k[X] = ∪i Vi .
Solution. Take f ∈ k[X]. If a∗ : k[X] → k[G]⊗k[X] maps f to i hi ⊗fi ,
P
then gf = i hi (g −1 )fi , which implies (i) and (ii). Now (ii) is a general
P
fact on countably dimensional locally finite modules.

9
Lie algebra of an algebraic group
9.1 Definitions
Let G be an algebraic group and A = k[G]. We will consider the Lie
algebra Der(A) of k-derivations A → A with respect to the bracket
[δ1 , δ2 ] = δ1 ◦ δ2 − δ2 ◦ δ1 . A derivation δ ∈ Der(A) is called left-invariant
if it commutes with left translations, i.e. δ ◦ λx = λx ◦ δ for all x ∈ G.
The left invariant derivations of A form a Lie subalgebra of Der(A),
called the Lie algebra of G and denoted L(G). (Using right invariant
derivations here would lead to an isomorphic object).
Let us denote by g the tangent space Te G. We claim that g can
be naturally identified with L(G) as vector spaces. Recall that Te G
can be defined as the the derivations of A at e. Define a k-linear map
θ : L(G) → g by
(θδ)(f ) = (δf )(e) (δ ∈ L(G), f ∈ A).
We claim that θ is an isomorphism of vector spaces. In order to prove

this we construct the inverse map η : g → L(g) sending a tangent vector
X to a derivation ∗X called right convolution by X and defined by
(f ∗ X)(x) = X(λx−1 f ) (x ∈ G, f ∈ A).
It is a straightforward check that ∗X is indeed a left invariant derivation

of A and that η is k-linear. Finally, η is inverse to θ:
(f ∗ θ(δ))(x) = θ(δ)(λx−1 f ) = δ(λx−1 f )(e) = λx−1 (δf )(e) = (δf )(x),

θ(∗X)(f ) = (f ∗ X)(e) = X(λe−1 f ) = X(f )
(for X ∈ g, δ ∈ L(G), f ∈ A, x ∈ G).

From now on we are going to identify L(G) with g via the isomor-
phisms θ and η. For example, g is a Lie algebra with respect to the
105
106 Lie algebra of an algebraic group
bracket defined as follows:
[X, Y ](f ) = ((f ∗ Y ) ∗ X − (f ∗ X) ∗ Y )(e) = X(f ∗ Y ) − Y (f ∗ X).
We give another definition of [X, Y ] in terms of the coproduct ∆. Define
X · Y : A → k, f 7→ (X ⊗ Y ) ◦ ∆(f ).
fi ⊗ fi0 , then
P
If ∆(f ) = i
X
f ∗X = fi X(fi0 ),
i
whence it is easy to see that (X · Y )(f ) = ((f ∗ Y ) ∗ X))(e). So
[X, Y ] = X · Y − Y · X.
This definition of the bracket makes the following easy to check:
Theorem 9.1.1 If ϕ : G → G0 is a homomorphism of algebraic groups,

then dϕ : g → g0 is a homomorphism of Lie algebras.
If H is a closed subgroup of an algebraic group G, the inclusion η :

H → G is an isomorphism onto a closed subgroup, with η ∗ : k[G] →
k[H] = k[G]/I being the natural projection. Therefore, dη identifies
h with the Lie subalgebra of g consisting of those X ∈ g for which
X(I) = 0. We will always identify h with a Lie subalgebra of g in this
way. Now, let ϕ : G → G0 be a morphism of algebraic groups, H 0 < G0
is a closed subgroup, and ϕ(H) ⊂ H 0 . Then ϕ|H can be considered
as a morphism H → H 0 , so its differential d(ϕ|H) is a Lie algebra
homomorphism h → h0 . It follows from the definitions that
(dϕ)|h : h → h0 = d(ϕ|H). (9.1)
Lemma 9.1.2 Let H be a closed subgroup of an algebraic group G and

I = I(H) / k[G]. Then h = {X ∈ g | I ∗ X ⊂ I}.
Proof If f ∈ I, X ∈ h, and x ∈ H, then (f ∗ X)(x) = X(λx−1 f ) = 0

since λx−1 f ∈ I. Conversely, if I ∗ X ⊂ I and f ∈ I, then (f ∗ X)(e) =
X(λe−1 f ) = X(f ) = 0, forcing X ∈ h.
Lemma 9.1.3 Let ρ : G → GL(V ) be a rational representation and dρ :

g → gl(V ) be the corresponding Lie algebra representation. If W ⊂ V is
a G-invariant subspace then W is also g-invariant.
9.2 Examples 107
Proof If we extend a basis
of W to a basis of V , then the matrix of
∗ ∗
any ρ(x) has the form , and so the matrix of any dρ(X) has the
0 ∗
same form.
9.2 Examples
Example 9.2.1 If G = Ga , then g is 1-dimensional, so its bracket is

trivial.
Example 9.2.2 Let G = GLn . Its tangent space at e has as a basis

the set of partial derivatives ∂T∂ij |e (evaluated at e). The coordinates xij
with repsect to this basis can be arranged in a square matrix. So we can
think of tangent vectors X as square matrices (xij ), where xij = X(Tij ).
P
With this convention (X · Y )(Tij ) = l xil ylj . In other words, X · Y is
the usual matrix product XY . Thus g = gln (k).
Example 9.2.3 The Lie algebra sln (k) of SLn < GLn consists of all
aij ∂T∂ij |e be
P
matrices in gln (k) of trace 0. Indeed, let X = (aij ) =
a tangent vector. Then X ∈ sln (k) if and only if X(det) = 0, which is
equivalent to tr X = 0.
Example 9.2.4 The group Sp2n < GL2n is Z(xt Jx−x) (4n2 polynomial
equations written as one matrix equation). So the Lie algebra sp2n (k)
consists of all matrices X ∈ gl2n (k) with X(xt Jx − x) = 0. This is
equivalent to X t J + JX = 0 (compute!). Compute dim sp2n (k).
Example 9.2.5 The group On < GL2n is Z(xxt − 1) (n2 polynomial

equations written as one matrix equation). So the Lie algebra son (k)
consists of all matrices X ∈ gln (k) with X + X t = 0.
Example 9.2.6 The Lie algebra u of the subgroup Un < GLn of upper
unitriangular matrices consists of all strictly upper triangular matrices
in gln (k).
Lemma 9.2.7 Let G be an algebraic group with product µ : G × G → G

and inverse ι : G → G. Then for all X, Y ∈ g:
(i) dµ(e,e) (X, Y ) = X + Y ;
(ii) dιe (X) = −X;
Proof Let (X, Y ) ∈ g ⊕ g = T(e,e) G × G, and Z := dµ(e,e) (X, Y ). If f ∈
k[G] and ∆(f ) = i fi ⊗ fi0 , then Z(f ) = i (X(fi )fi0 (e) + fi (e)Y (fi0 )),
P P
cf. the proof of Proposition 6.1.5. On the other hand, we have

X X
f= fi (e)fi0 = fi0 (e)fi .
i i
(you should have checked that when you checked the axioms of Hopf
algebra for k[G], but it’s not too late now). So Z(f ) = (X + Y )(f ),
giving (i).
Consider the composite G → G × G → G, g 7→ (g, ι(g)) 7→ gι(g) = e.
The composite is a constant function, so its differential is zero. But
the differential of a composite is the composite of the differentials, so
applying (i), we have 0 = d ide +dιe = id +dιe , whence (ii).
Lemma 9.2.8 Let E ⊂ k[G] be a finite dimensional subrepresentation

of the (right) regular representation ρ of G, and ψ : G → GL(E) be the
restriction of ρ to E. Then dψ(X)(f ) = f ∗ X for f ∈ E.
P
Proof Pick a basis {f1 , . . . , fn } of E. Let ∆(fi ) = j fj ⊗ mij , see
P
(8.1). Then ρx (fi ) = j mij (x)fj . So the matrix of ψ(x) in our basis
is (mij (x)). Note, moreover, that
X
λx−1 fi = fj (x)mij . (9.2)
j
Now, let X ∈ g. By definition, the (i, j)entry of the matrix dψ(X) is

X(ψ ∗ (Tij )) = X(mij ). On the other hand, using (9.2), we get
X
(fi ∗ X)(x) = X(λx−1 fi ) = fj (x)X(mij ),
j
which completes the proof.
9.3 Ad and ad
Fix x ∈ G. Let Int x : G → G, y 7→ xyx−1 . The differential d(Int x)e is
a Lie algebra automorphism denoted
Ad x : g → g.
The image of Ad is a (closed connected) subgroup of GL(g)) denoted

Ad G.
9.3 Ad and ad 109
Example 9.3.1 Let G = GLn . Then Ad x(X) = xXx−1 (for X ∈ g =
gln (k)). Hence for any closed subgroup H < G, its Lie algebra h, and
x ∈ H, Ad x : h → h is conjugation by x too.
For the proof, let us compute (Int x)∗ (Tij ):
X
(Int x)∗ (Tij )(g) = Tij (xgx−1 ) = xik Tkl (g)(x−1 )lj .
k,l
Hence
X
(Int x)∗ (Tij ) = xik (x−1 )lj Tkl .
k,l
Now, the ij-entry of Ad x(X) is

X
Ad x(X)(Tij ) = X((Int x)∗ (Tij )) = xik (x−1 )lj X(Tkl ),
k,l
which is the ij-entry of xXx−1 .
Theorem 9.3.2 Ad is a rational representation of G in (the vector

space) g (called the adjoint representation of G).
Proof Embed G as a closed subgroup of some GLn . Then by Exam-

ple 9.3.1, Ad x is a conjugation by x, which implies that Ad : G → GL(g)
is a morphism of varieties.
Let ad : g → gl(g) be the adjoint representation of Lie algebra, i.e.
ad X(Y ) = [X, Y ] (X, Y ∈ g).
Theorem 9.3.3 The differential of Ad is ad.
Proof Using embedding of G into some GLn and (9.1), it suffices to

check the result for G = GLn . Note that Ad x is the image of x under
the map
(1,ι) σ×τ µ
G −→ G × G −→ GL(g) × GL(g) −→ GL(g),
where σ(x) (resp. τ (x)) is the left (resp. right) multiplication by x

in g. Since the entries of σ(x) and τ (x) are linear polynomials in the
entries of x, it follows that dσ(X) (resp. dτ (X)) is a left (resp. right)
multiplication by X. Now the result follows from Lemma 9.2.7.
9.4 Properties of subgroups and subalgebras
Lemma 9.4.1 If H is a closed normal subgroup of an algebraic group

G, then h is an ideal g.
Proof We have Int x stabilizes H for all x ∈ G. Hence Ad x stabilizes h

for all x ∈ G. If we extend a basis
of hto a basis of g, then the matrix
∗ ∗
of Ad x therefore has the form (x ∈ G), and so the matrix of
0 ∗
d(Ad)(X) = ad X has the same form (X ∈ g).
Lemma 9.4.2 If H is a closed subgroup of an algebraic group G and

N = NG (H), then n ⊂ ng (h).
Proof Note that N is closed in view of Lemma 8.3.1(i). Applying

Lemma 9.4.1 to the normal subgroup H of N , we see that h is an ideal
of n, i.e. n normalizes h.
For x ∈ G denote
γx : G → G, y 7→ yxy −1 x−1 .
Lemma 9.4.3 (dγx )e (X) = X − Ad x(X).
Proof Consider first the morphism ψ : G → G, y 7→ xy −1 x−1 . As

ψ = Int x ◦ ι, we have
dψe (X) = d(Int x) ◦ dιe (X) = Ad x(−X) = − Ad(X).
Now γx can be realizete as the composite
(1,ψ) µ
G −→ G × G −→ G.
So (dγx )e (X) = dµ(e,e) (X, dψe (X)) = X − Ad x(X).
Lemma 9.4.4 Let x ∈ G. Then L(CG (x)) ⊂ cg (x) := {X ∈ g |

Ad x(X) = X}. If G = GLn , then equality holds.
Proof Note that the Lie algebra L(CG (x)) of the fiber γx−1 (e) = CG (x)
maps to zero under the map (dγx )e . Now use Lemma 9.4.3.
In case of GLn the fixed points of Ad x in g are just the matrices
commuting with x, so CG (x) is a principal open set in cg (x), containing
e, which implies the result.
9.5 Automorphisms and derivations 111
Lemma 9.4.5 Let ρ : G → GL(V ) be a rational representation, and
dρ : g → gl(V ) be the corresponding representation of the Lie algebra. If
v ∈ V , let CG (v) = {x ∈ G | xv = v} and cg (v) := {X ∈ g | Xv = 0}.
Then L(CG (v)) ⊂ cg (v).
Proof Note that x 7→ xv is a morphism G → V constant on CG (v), so

dρe is zero on the Lie algebra L(CG (v)).
Lemma 9.4.6 Let A and B be closed subgroups of G, and let C be the

closure of the subgroup C = (A, B) generated by the commutators. The
its Lie algebra c contains all elements of the form [X, Y ], Y − Ad x(Y ),
X − Ad y(X) (x ∈ A, X ∈ a, y ∈ B, Y ∈ b). In particular, if H is the
closure of (G, G), then h ⊃ [g, g].
Proof For x ∈ A, γx maps A to C, so the differential 1 − Ad x maps

b to c. This yields all elements of the second type listed, and similarly
for the third type. Next for X ∈ a consider the morphism ϕ : B → c
defined by ϕ(y) = X − Ad y(X). Since ϕ maps e to 0, we have dϕe (Y ) =
− ad Y (X) = −[Y, X] = [X, Y ].
Remark 9.4.7 Inclusions in Lemmas 9.4.2, 9.4.4, 9.4.5, and 9.4.6 can
be proper in positive characteristic and are equalities in characteristic
0.
9.5 Automorphisms and derivations
Lemma 9.5.1 Let V and W be rational G-modules. Then

(i) g acts on V ∗ by the rule Xf (v) = −f (Xv) for f ∈ V ∗ , v ∈ V, X ∈
g.
(ii) g acts on V ⊗ W by the rule X(v ⊗ w) = (Xv) ⊗ W + v ⊗ (Xw)
for v ∈ V, w ∈ W, X ∈ g.
Proof (i) We fix a basis of V and write the action of x ∈ G as a matrix.

Then the matrix of x acting on the dual basis of V ∗ is the transpose
inverse matrix. We know that the differential of x 7→ x−1 is X 7→ −X,
while the map x 7→ xt of GLn has the differential X 7→ X t on gln . This
implies the result.
(ii) Fix bases {v1 , . . . , vn } of V and {w1 , . . . , wm } of W , and let ρ1 :
G → GLn , ρ2 : G → GLm be the corresponding matrix representations.
If ρ1 (x) = (aij ) and ρ2 (x) = (brs ), then the matrix (ρ1 ⊗ ρ2 )(x) has
entry air bjs in the (i, j)row and (r, s)column. So the representation
G → GLmn factors as the composite of two morphisms
(ρ1 ,ρ2 )
G −→ GLn × GLm → GLmn ,
where the second map is given in coordinates via Zij,rs = Xir Yjs . It is
easy to compute the differential (at (e, e)) of the second morphism—it
maps a pair of matrices ((cij ), (drs )) ∈ gln ⊕ glm to the matrix whose
entry in row (i, j) and column (r, s) is δjs cir + δir djs . This implies the
rule asserted in (ii).
Corollary 9.5.2 Let B be a finite dimensional k-algebra (not necessarily

associative), and let G be a closed subgroup of GL(B), consisting of
algebra automorphisms. Then g consists of derivations of B.
Proof Let t ∈ B∗ ⊗ B ∗ ⊗ B = Homk (B ⊗ B, B) be the multiplication on

B. Note that x ∈ GL(B) is an automorphism of B if and only if t is
an invariant of x. So t is an invariant of G, whence it is an invariant
of g, see Lemma 9.4.5. This is equivalent to the fact that g consists of
derivations of B.
9.6 Problems
Problem 9.6.1 Let H be a closed subgroup of G = GL(V ), h ⊂ gl(V )

be its Lie algebra, v ∈ V , and W ⊂ V be a vector subspace.
(i) If H leaves W stable, then so does h. Is the converse true?
(ii) If H leaves v stable, then h kills v. Is the converse true?
(iii) Set GW := {x ∈ G | x(W ) ⊂ W }, gW := {X ∈ g | X(W ) ⊂ W }.
Then L(GW ) = gW . (Hint: L(GW ) ⊂ gW by (i). Now, use ex-
plicit descriptions of GW and gW using matrices and dimensions).
(iv) Set Gv := {x ∈ G | xv = v}, gv := {X ∈ g | Xv = 0}. Then
L(Gv ) = gv .
Problem 9.6.2 Prove that Z(G) ⊂ ker Ad.
 Let chark = p > 0 and G ⊂ GL3 consist of all matrices

Problem 9.6.3
a 0 0
of the form 0 ap b  with a 6= 0. Observe that g consists of all
0 0 1
9.6 Problems 113
 
a 0 0
matrices 0 0 b  and is commutative. Moreover, {e} = Z(G) (
0 0 0
ker Ad ( G.
Problem 9.6.4 Let char k = 2, G = SL2 , and B the group of all upper
triangular matrices in G. Then NG (B) = B, whereas ng (b) = g.
Problem 9.6.5 Let H < G be a closed subgroup and x ∈ G. Then

Ad x(h) = L(Int x(H)).
Problem 9.6.6 Define P GLn := Ad GLn and P SLn := Ad SLn . The

centers of GLn and SLn consist of all scalar matrices contained in
these groups. As abstract groups P GLn ∼ = GLn /Z(GLn ) and P SLn ∼ =
SLn /Z(SLn ). If the characteristic p of k divides n, then Z(SLn ) = {1},
but SLn is not isomorphic to P SLn as algebraic groups!!!!
Problem 9.6.7 If char k = p > 0 and X is a left invariant derivation

of k[G], then X p is also a left invariant derivation of k[G]. This gives
an extra operation on g, called pth power operation, which makes g into
a restricted Lie algebra. (One needs to check a number of axioms here,
but never mind...) Compute the pth power operation for G = Ga and
G = GLn .
10
Quotients
The main problem addressed in this chapter is as follows: given an

algebraic group G and a closed subgroup H, how to endow the quotient
G/H with a ‘reasonable’ structure of algebraic variety?
10.1 Construction
We start with a linear algebra lemma.
Lemma 10.1.1 Let M be a d-dimensional subspace of a vector space

W , x ∈ GL(W ), X ∈ gl(W ). Then L := Λd M can be considered as a
line in Λd W .
(i) xL = L if and only if xM = M .
(ii) XL ⊂ L if and only if XM ⊂ M .
Proof Exercise or read it in Humpreys.
Theorem 10.1.2 (Chevalley) Let G be an algebraic group, H < G a

closed subgroup. Then there is a rational representation ϕ : G → GL(V )
and a 1-dimensional subspace L of V such that H = {x ∈ G | ϕ(x)L =
L} and h = {X ∈ g | dϕ(X)L ⊂ L}.
Proof Let I = I(H) / k[G]. Let W ⊂ k[G] be a finite dimensional sub-

space invariant with respect to all ρx and containing a (finite) generating
set for I, see Lemma 8.4.2. Let M = W ∩ I (so M generates I). Note
that H = {x ∈ G | ρx I = I}, so M is stable under all ρy for y ∈ H. It
follows from Lemmas 9.2.8 and 9.1.3 that M is stable under all ∗Y for
Y ∈ h.
114
10.2 Quotients 115
We claim that H = {x ∈ G | ρx M = M } and h = {X ∈ g | M ∗ X ⊂
M }. Indeed, if ρx M = M , then we have
ρx I = ρx (M A) = ρx (M )ρx (A) = M A = I,
forcing x ∈ H. If M ∗ X ⊂ M then the product rule implies
I ∗ X = (M A) ∗ X ⊂ (M ∗ X)A + M (A ∗ X) ⊂ M A = I,
forcing X ∈ h by Lemma 9.1.2.
Finally, pass to Λd W where d = dim M , take ϕ to be the dth exterior
power of the representation constructed above, and use Lemma 10.1.1.
Corollary 10.1.3 Let H be a closed subgroup of a connected algebraic

group G. Then there exists a quasi-projective variety X that G acts
transitively on and a point x ∈ X such that
(i) Gx = H;
(ii) the orbit map ψ : G → X, g 7→ gx is separable;
(iii) the fibers of ψ are the cosets gH of H in G.
Proof Let V and L = hvi ⊂ V be as in the theorem. Take X to be the

G-orbit Ghvi in P(V ) and x = hvi. This is open in its closure, hence it
is a quasi-projective variety. By the theorem, H = Gx , and now (iii) is
also clear.
Finally note that the tangent space to P(V ) at x can be canonically
identified with V /hvi, and the tangent space to X at x is a subspace
of V /hvi. The differential dψe maps Y ∈ g to Y v + hvi. Now, by the
theorem, the kernel of the differential is h. So
dim ker dψe = dim h = dim H = dim G − dim X.
Hence dψe is onto by dimension, and ψ is separable in view of Theo-
rem 6.7.1.
10.2 Quotients
In this section we will assume that G is a connected algebraic group
and H < G a closed subgroup. (The assumption that G is connected is
not essential, but we do not want to deal with necessary modifications
needed in the non-connected case).
A Chevalley quotient of G by H is a variety X together with a surjec-
tive separable morphism π : G → X such that the fibers of π are exactly
116 Quotients
cosets of H in G. By Corollary 10.1.3 Chevalley quotients exist, but it
is not clear if they are unique up to isomorphism.
A categorical quotient of G by H is a variety X together with a mor-
phism π : G → X that is constant on all cosets of H in G with the
following universal property: given any other variety Y and a morphism
ϕ : G → X that is constant on all cosets of H in G there is a unique
morphism ϕ̄ : X → Y such that ϕ = ϕ̄ ◦ π. Now, it is clear that categor-
ical quotients are unique up to unique isomorphism, but it is not clear
if they exist.
Our goal is to prove that Chevalley quotients are categorical quo-
tients. This will prove that categorical quotients exist and that Cheval-
ley quotients are unique. So we need to take a Chevalley quotient (X, π)
and check that it has the right universal property. Given a morphism
ϕ : G → Y constant on cosets, there is a unique map of sets X → Y
such that ϕ = ϕ̄ ◦ π, since fibers of π are exactly the cosets. But it is
very difficult to prove from this point of view that ϕ is a morphism of
varieties. So we proceed rather differently.
Theorem 10.2.1 Chevalley quotients are categorical quotients.
Proof Step 1. Let us try to construct a categorical quotient not in

the category of varieties but in the more general category of geometric
spaces. Define G/H to be the set of cosets of H in G. Let π : G → G/H
be the map x 7→ xH. Give G/H the structure of topological space by
declaring U ⊂ G/H to be open if and only if π −1 (U ) is open. Next
define a sheaf O of functions on G/H: if U ⊂ G/H is open, let O(U )
consist of all functions f on U such that f ◦ π ∈ OG (π −1 (U )). (Check
the sheaf axioms!)
In order to check the universal property, let ψ : G → Y be a morphism
of geometric spaces constant on the cosets of H in G. We get the induced
map of sets ψ̄ : G/H → Y, xH 7→ ψ(x). We claim that ψ is a morphism
of geometric spaces. For continuity, take an open subset V ⊂ Y , and
note that U := ψ̄ −1 (V ) is open in G/H, as π −1 (ψ̄ −1 (V )) = ψ −1 (V ) is
open in G. Finally, take f ∈ OY (V ) and show that ψ̄ ∗ (f ) ∈ OG/H (U ).
By definition, we just need to check that π ∗ (ψ̄ ∗ (f )) ∈ OG (ψ −1 (V )).
But π ∗ (ψ̄ ∗ (f )) = ψ ∗ f ∈ OG (ψ −1 (V )), as ψ is a morphism of geometric
spaces.
Step 2. Now, let (G/H, π) be as in step 1, and let (X, ψ) be a Cheval-
ley quotient. Using the universal property established above, we get a
unique G-equivariant morphism ψ̄ : G/H → X such that ψ = ψ̄ ◦ π,
10.2 Quotients 117
i.e. ψ̄(xH) = ψ(x). We will prove that ψ̄ is an isomorphism of geo-
metric spaces, which will imply that G/H is a variety and that X is a
categorical quotient.
First of all, it is clear that ψ̄ is bijective. Moreover, by Lemma 8.3.7,
the map ψ is open (and continuous), which implies that ψ̄ is a home-
omorphism. In order to finish the proof, take an open subset U ⊂ X,
a function f ∈ OG (ψ −1 (U )) constant on the cosets, and prove that
f = ψ ∗ (g) for some g ∈ OX (U ). For simplicity we consider the case
U = X when ψ −1 (U ) = G. The argument for the general case is simi-
lar.
We show first that there exists a rational function g with the required
property, i.e. f = ψ ∗ (g) in k(G). Consider the morphisms
ϕ π
G −→ X × A1 −→
1
X,
where ϕ = (ψ, f ). The composite is just ψ. If Y is the closure in X × A1

of ϕ(G), then Y is irreducible, and π1 induces a surjective morphism
η : Y → X. Since ψ is separable, so is η (use ψ ∗ = ϕ∗ ◦ η ∗ ).
Now, ϕ(G) contains a dense open subset of Y , see Problem 5.5.3.
Since f is constant on fibers of ψ, the restriction of η to this open set
is injective, as well as dominant and separable. By Theorem 5.4.3, η ∗
maps k(X) isomorphically onto k(Y ). But π2 : X × A1 → A1 induces
on Y a morphism g : Y → A1 , i.e. a regular function, in particular a
rational function. So there exists h ∈ k(X) for which g = η ∗ h. Finally,
notice that ϕ∗ g = ϕ∗ η ∗ h = ψ ∗ h agrees everywhere on G with f . So
f = ψ ∗ h, as desired.
Next we want to show that the rational function h ∈ k(X) just con-
structed is actually a regular function on Y . Since all points of X are
simple, Theorem 6.3.2 shows that unless h is everywhere defined on X,
1/h is defined and is equal to 0 at some point. But then ψ ∗ (1/h) = 1/f
must also take the value zero, which is absurd since f ∈ k[G].
We will denote by G/H the categorical quotient of G by the closed

subgroup H. We now know that the categorical quotient exists and
is unique up to a unique isomorphism. We also know that G/H is a
quasi-projective variety and π : G → G/H is separable. Also note that
TeH (G/H) ∼
= g/h.
Indeed, the separability of π implies that dπeH is surjective, and it con-

tains h in its kernel. Now use dimensions.
118 Quotients
Example 10.2.2 Let G = GL(V ), {v1 , . . . , vn } be a basis of G, and
G = Ghv1 i . Then X = Ghv1 i = P(V ) is a Chevalley quotient of G by
H, see the proof of Corollary 10.1.3. So G/H ∼ = P(V ). Similarly, let
P = Ghv1 ,...,vd i . Then G/P = Gd (V ). Finally, G/B ∼
∼ = F, where B is
the stabilizer of a standard flag.
In the examples above quotients are projective varieties. Let K :=
Gv1 . Then G/K ∼ = An \ {0}. This is neither affine nor projective (unless
n = 1).
Example 10.2.3 Let G = GLn and H = On = {g ∈ GLn | g t g = I}

(assuming char k 6= 2). Let S be the set of all n × n symmetric matrices,
and affine variety of dimension n(n + 1)/2. Let S × be the invertible
matrices in S, a principal open subset of S, hence also affine of dimension
n(n + 1)/2.
Let G act on S by g · x = g t xg. Then GI = On , and the action
is transitive, as by linear algebra all non-degenerate symmetric bilinear
forms are equivalent. To prove that GLn /On ∼ = S × , we just need to
prove that the orbit map G → S × , g 7→ g t g is separable. Its differential
is the map X 7→ X t + X. The tangent space to S × at I can be identified
with S. Clearly any symmetric matrix can be written in the form X t +X
(characteristic is not 2!).
Thus GLn /On ∼ = S × , which is an affine variety.
10.3 Normal subgroups

Let G be an algebraic group. A character of G is a homomorphism
χ : G → Gm of algebraic groups. We write X(G) for the set of all
characters of G. It has a natural structure of an abelian group:
(χ + ψ)(g) = χ(g)ψ(g).
Let V be a rational G-module. For χ ∈ X(G), let
Vχ := {v ∈ V | gv = χ(g)v for all g ∈ G}.

P
It is easy to see that χ∈X(G) Vχ = ⊕χ∈X(G) Vχ , see Problem 10.4.5. On
P
the other hand, it is usually not true that V = χ∈X(G) Vχ . But there is
one important case when it is the case. This is when G = Dn ∼ = (Gm )n ,
the group of all diagonal matrices in GLn . This will be established later.
Now, let N be a closed normal subgroup of G, and V be a rational
G-module. If χ ∈ X(N ), then for any g ∈ G we have gVχ ⊂ Vχ0 for
10.3 Normal subgroups 119
χ0 = gχ ∈ X(N ). Here gχ(h) := χ(g −1 hg). Indeed, let v ∈ Vχ and
h ∈ N . Then hgv = gg −1 hgv = χ(g −1 hg)gv.
Theorem 10.3.1 Let G be an algebraic group and N ⊂ G be a closed

normal subgroup. Then the variety G/N is affine, and G/N × G/N →
G/N, (g1 N, g2 N ) 7→ g1 g2 N , G/N → G/N, gN 7→ g −1 N are morphisms
of varieties.
Proof Let us show first that (g1 N, g2 N ) 7→ g1 g2 N is a morphism. The

map G × G → G/N, (g1 , g2 ) 7→ g1 g2 N is a morphism that is constant on
cosets of N ×N . Hence by the universal property of the quotients, we get
induced a unique morphism G/N × G/N ∼ = (G × G)/(N × N ) → G/N ,
see Problem 10.4.1. The proof that gN 7→ g −1 N is a morphism is similar
(but easier).
By Chevalley’s theorem, we can find a rational representation ρ : G →
GL(V ) and v ∈ V such that H = Ghvi , and h is the stabilizer of hvi in
g. Let V 0 = ⊕χ∈X(N ) Vχ . Note that v ∈ V 0 and V 0 is G-invariant, so we
may assume that V = V 0 .
Now, let W = {f ∈ End (V ) | f (Vχ ) ⊂ Vχ for all χ ∈ X(H)}. Define
a morphism of algebraic groups ψ : G → GL(W ), where
ψ(g)f = ρ(g)f ρ(g)−1 (g ∈ G, f ∈ W ).
Let us compute the kernel of ψ: if ψ(g) = id, then ρ(g) stabilizes each
Vχ and commutes with End (Vχ ), hence by Schur’s lemma ρ(g) acts as
scalars on each Vχ . Hence g stabilizes hvi, so g ∈ H. Conversely, if H
acts as a scalar on each Vχ , then H ⊂ ker ψ.
Note that the image of ψ is a closed—hence affine—subgroup of GL(W ).
To show that this is a Chevalley quotient we just need to prove that
ψ is separable. For this we show that dψ is onto, or equivalently by
dimensions that ker dψ ⊂ h. Let X ∈ ker dψ. Then dψ(X)(f ) =
dρ(X)f − f dρ(X) = 0, so dρ(X) commutes with all f ∈ W . This
implies that dρ(X) acts as a scalar on all Vχ ’s, in particular, it stabilizes
hvi, hence X ∈ h.
Corollary 10.3.2 Suppose that ϕ : G → H be a separable surjective

morphism of algebraic groups and N = ker ϕ. Then ϕ induces an iso-
morphism G/N ∼ = H.
Note in characteristic 0 the separability is automatic. On the other

hand, let G = H = GLn , and let ϕ be the Frobenius homomorphism
120 Quotients
given by raising matrix entries to the pth power. This is a morphism
of algebraic groups and an isomorphism of abstract groups, but the
differential dϕ is the zero map. So ϕ is definitely not an isomorphism of
algebraic groups.
10.4 Problems
Problem 10.4.1 Let H1 < G1 , H2 < G2 be closed subgroups of con-

nected algebraic groups. Prove that (G1 × G2 )/(H1 × H2 ) ∼
= G1 /H1 ×
G2 /H2 .
Problem 10.4.2 Prove that GL2n /Sp2n is isomorphic to the affine vari-
ety of all invertible 2n × 2n-skew symmetric matrices. (In characteristic
2 a skew symmetric matrix means a symmetric matrix with zeros on the
main diagonal).
Problem 10.4.3 Prove that X(SLn ) = {0}, X(Ga ) = {0}, X(Gm ) ∼

=
Z, X(GLn ) = Z.
Problem 10.4.4 Prove that X(G × H) ∼

= X(G) ⊕ X(H).
P
Problem 10.4.5 Prove that χ∈X(G) Vχ = ⊕χ∈X(G) Vχ
Problem 10.4.6 Let A, B ⊂ G be closed subgroups. Prove that a ∩ b =

L(A ∩ B) if and only if the restriction to A of the canonical morphism
π : G → G/B is again separable. (Hint: consult Theorem 12.1.1.)
Problem 10.4.7 Let H be a closed subgroup of a connected algebraic

group G. Then H acts naturally on k(G) as a group of automorphisms,
and k(G/H) ∼
= k(G)H .
Problem 10.4.8 Compute the dimension of the flag variety.

11
Semisimple and unipotent elements
11.1 Jordan-Chevalley decomposition

The following result about the additive Jordan decomposition is well
known:
Lemma 11.1.1 Let V be a finite dimensional k-vector space and X ∈

End V .
(i) There exist unique Xs , Xn ∈ End (V ) satisfying the conditions
X = Xs + Xn , Xs is semisimple, Xn is nilpotent, and Xs Xn =
Xn Xs .
(ii) There exist polynomials p(T ), q(T ) without constant term such
that Xs = p(X), Xn = q(X). In particular Xs , Xn commute with
any endomorphism of V which commutes with X.
(iii) If A ⊂ B ⊂ V are subspaces and X maps B to A, then so do Xs
and Xn .
(iv) If XY = Y X for Y ∈ End V then (X + Y )s = Xs + Ys and
(X + Y )n = Xn + Yn .
(v) If ϕ : V → W is a linear map and Y ∈ End W such that Y ◦ ϕ =
ϕ ◦ X, then Ys ◦ ϕ = ϕ ◦ Xs and Yn ◦ ϕ = ϕ ◦ Xn .
An element x ∈ End V is called unipotent if it is the sum of idV and

a nilpotent element, or, equivalently, if the only eigenvalue of x is 1.
In characteristic p an element x ∈ End V is unipotent if and only if
N
xp = 0 for some N . The additive Jordan decomposition implies the
multiplicative Jordan decomposition:
Lemma 11.1.2 Let V be a finite dimensional k-vector space and x ∈

GL(V ).
121
122 Semisimple and unipotent elements
(i) There exist unique xs , xu ∈ End (V ) satisfying the conditions x =
xs xu , xs is semisimple, xu is unipotent, and xs xu = xu xs .
(ii) xs , xu commute with any endomorphism of V which commutes
with x.
(iii) If A ⊂ V is a subspaces stable under x, then A is stable under xs
and xu .
(iv) If xy = yx for y ∈ GL(V ) then (xy)s = xs ys and (xy)u = xu yu .
(v) If ϕ : V → W is a linear map and y ∈ End W such that y ◦ ϕ =
ϕ ◦ x, then ys ◦ ϕ = ϕ ◦ xs and yu ◦ ϕ = ϕ ◦ xu .
We leave the following as an exercise:
Lemma 11.1.3 Let x = xs xu and y = ys yu be Jordan decompositions of

x ∈ GL(V ) and y ∈ GL(W ). Then x⊕y = (xs ⊕ys )(xu ⊕yu ) and x⊗y =
(xs ⊗ ys )(xu ⊗ yu ) are Jordan decompositions of x ⊕ y ∈ GL(V ⊕ W )
and x ⊗ y ∈ GL(V ⊗ W ).
Theory of Jordan decompositions generalize to infinite dimensional

vector spaces V providing we restrict our attention to locally finite endo-
morphisms x, i.e. endomorphisms such that any v ∈ V belongs to a finite
dimensional x-invariant subspace. A locally finite endomorphism x of V
is semisimple if its restriction to every finite dimensional x-invariant sub-
space of V is semisimple. Nilpotent and unipotent are defined similarly.
For a general locally finite x ∈ End V we have its Jordan decompositions
x = xs + xn and x = xs xu , with all the properties of the finite dimen-
sional case holding. To define xs , take v ∈ V , find a finite dimensional
x-invariant subspace W containing v and define xs (v) = (x|W )s (v). The
fact that this is well-defined follows from the uniqueness statement in
the finite dimensional Jordan decomposition. The elements xn and xu
are defined similarly.
Theorem 11.1.4 For any x ∈ G, there are unique elements xs , xu ∈ G

such that (ρx )s = ρxs , (ρx )u = ρxu , and x = xs xu = xu xs . Moreover, if
ϕ : G → H is a morphism of algebraic groups, then ϕ(xs ) = ϕ(x)s and
ϕ(xu ) = ϕ(x)u .
Proof Let m : k[G] ⊗ k[G] → k[G] be the algebra multiplication. We

have
m ◦ (ρx ⊗ ρx ) = ρx ◦ m.
11.1 Jordan-Chevalley decomposition 123
Hence by Lemmas 11.1.2(v) and 11.1.3,
m ◦ ((ρx )s ⊗ (ρx )s ) = (ρx )s ◦ m,
i.e. (ρx )s respects the multiplication on k[G]. Also ρx (1) = 1 implies

(ρx )s (1) = 1 by the properties of Jordan decomposition. Thus (ρx )s
is an automorphism of k[G]. Hence ξ : f 7→ ((ρx )s f )(e) is an algebra
homomorphism k[G] → k. So there is a point xs ∈ G with ξ(f ) = f (xs ).
To prove that (ρx )s and ρxs are the same note that λy and ρx commute
for all y, so λy and (ρx )s commute too. Now,
((ρx )s f )(y) = (λy−1 (ρx )s f )(e) = ((ρx )s λy−1 f )(e)

= (λy−1 f )(xs ) = f (yxs ) = (ρxs f )(y).
Similarly we find xu such that (ρx )u = ρxu . But the right regular repre-
sentation is faithful, so ρx = ρxs ρxu = ρxu ρxs implies x = xs xu = xu xs .
Now, let x ∈ G and y = ϕ(x). It is easy to check that ϕ∗ ◦ρy = ρx ◦ϕ∗ .
Hence ϕ∗ ◦ (ρy )s = (ρx )s ◦ ϕ∗ . So ϕ∗ ◦ ρys = ρxs ◦ ϕ∗ . For any f ∈ k[H],
(ϕ∗ (ρys (f )))(e) = (ρys (f ))(ϕ(e)) = (ρys (f ))(e) = f (ys ).
This equals
(ρxs ◦ ϕ∗ )(f )(e) = (ϕ∗ f )(xs ) = f (ϕ(xs )).
We conclude that ϕ(xs ) = ys . The argument for the unipotent parts is

similar.
Remark 11.1.5 One can also prove the infinitesimal analogue of this
result: for any X ∈ g, there are unique elements Xs , Xn ∈ g such that
(∗X)s = ∗Xs , (∗X)u = ∗Xn , [Xs , Xn ] = 0, and X = Xs +Xn ; moreover,
if ϕ : G → H is a morphism of algebraic groups, then dϕ(Xs ) = dϕ(X)s
and dϕ(Xn ) = dϕ(X)n . See Humphreys for details.
Decompositions x = xs xu and X = Xs +Xn coming from the theorem

and the remark are refereed to as the abstract Jordan decompositions or
Jordan-Chevalley decompositions. If x = xs , we call x semisimple, and of
x = xu we call u unipotent. The set of all semisimple (resp. unipotent)
elements of G is denoted Gs (resp. Gu ).
Example 11.1.6 If x ∈ G = GLn , then xs is just the semisimple part

of x considered as an endomorphism of V = k m , and xu is the unipotent
part. To see this, let f ∈ V ∗ be a non-zero functional. For v ∈ V define
f˜(v) ∈ k[G] by f˜(v)(x) = f (xv). This gives an injective linear map
f˜ : V → k[G] which satisfies f˜(xv) = ρx f˜(v). Hence
f˜(xs v) = (ρx )s f˜(v) = ρxs f˜(v) = f˜(xs v),
where the first xs is the semisimple part of x in the old sense of linear
algebra, and the other two xs ’s refer to the semisimple part of x in the
abstract Jordan decomposition. This implies that the two are the same.
The argument for the unipotent parts and Lie algebras is similar.
For an arbitrary G, we can embed it as a closed subgroup of some
GL(V ). Then again, the abstract Jordan decompositions x = xs xu of x
as an element of G and as an endomorphism of V coincide.
11.2 Unipotent algebraic groups

An algebraic group is called unipotent if all of its elements are unipotent.
Theorem 11.2.1 Let G be a unipotent closed subgroup of GLn . Then

there is g ∈ GLn such that gGg −1 < Un .
Proof Let V = k n . It suffices to show that G fixes some flag in V . Using

induction on n we may assume that G does not stabilize any subspace
of V , i.e. G acts irreducibly on G. Then by Wedderburn theorem the
elements of G span the vector space End V . Since G is unipotent, all
elements of G have trace n. Hence 0 = tr(h − gh) = tr(1 − g)h for all
g, h ∈ G, hence for all g ∈ G and all h ∈ End V . Taking h to be various
matrix units, you now get that 1 − g = 0, i.e. G = {e}.
Corollary 11.2.2 Unipotent algebraic groups are nilpotent.
Theorem 11.2.3 (Rosenlicht) Let G be an unipotent algebraic group

acting on an algebraic variety X. Then all orbits of G on X are closed.
Proof Let O be an orbit. Replacing X by Ō, we may assume that O

is open dense in X. Let Y be its complement. Consider the action of
G on k[X] by translation of functions. This action is locally finite, see
Problem 8.5.9. Moreover, G stabilizes Y , so it leaves I(Y ) invariant. By
Theorem 11.2.1, there is a non-zero function f ∈ I(Y ) fixed by G. But
then f is constant on O. So, since O is dense, f is constant on X. This
shows that f is a non-zero scalar, hence I(Y ) = k[X] and Y = ∅.
11.3 Problems 125
Now let G be an arbitrary connected algebraic group. Suppose that
X, Y are two closed connected normal solvable subgroups of G. Then
XY is again a closed connected normal solvable subgroup of G. It follows
that G contains a unique maximal closed connected normal solvable
subgroup. This is called the radical of G and denoted R(G). Similarly
one defines the unipotent radical Ru (G) as the unique maximal closed
connected normal unipotent subgroup.
A connected algebraic group is called semisimple if R(G) = {e} and
reductive if Ru (G) = {e}. Unipotent groups are nilpotent, so semisimple
groups are reductive. There is a beautiful structure theory and classifi-
cation of reductive groups.
Lemma 11.2.4 If M ⊂ Mn (k) is a commuting set of matrices, then M

is triagonalizable. If M consists of the diagonalizable matrices, then M
is diagonalizable.
Proof Linear algebra. See Humpreys, 15.4.
Theorem 11.2.5 Let G be a commutative algebraic group. Then Gs

and Gu are closed subgroups of G, connected if G is, and the product
map ϕ : Gs × Gu → G is an isomorphism of algebraic groups.
Proof That Gs , Gu are subgroups follows from Lemma 11.1.2(iv). That

Gu is closed is Problem 11.3.1. Moreover, Theorem 11.1.4 implies that
ϕ is an isomorphism of abstract groups. Now embed G into some GLn .
Lemma 11.2.4 allows us to assume that G is a group of upper triangular
matrices and that Gs is a group of diagonal matrices. This implies that
Gs is also closed.
It has to be shown that the inverse map is a morphism or that the
maps x 7→ xs and x 7→ xu are morphisms. The second is if the first is,
as xu = x−1s x. Now xs is just the diagonal part of the matrix x (why?),
so x 7→ xs is a morphism. Now the connectedness of G also implies that
of Gs and Gu .
11.3 Problems
Problem 11.3.1 The set of all unipotent elements of G is closed.

Problem 11.3.2 Let B be a finite dimensional k-algebra. If x ∈ AutB,
then xs , xu ∈ AutB.
Problem 11.3.3 Let char k = 0. An element of GLn having finite order

must be semisimple.
Problem 11.3.4 Let G be a connected algebraic group of positive di-

mension. Prove that R(G) = {e} if and only if G has no closed connected
commutative normal subgroup. (Hint: see Example 8.2.10).
Problem 11.3.5 If char k = 0, then every unipotent subgroup of GLn

is connected.
Problem 11.3.6 If char k = 0 then 1-dimensional unipotent group is

isomorphic to Ga .
12
Characteristic 0 theory
Throughout this chapter we assume char k = 0.
12.1 Correspondence between groups and Lie algebras
Theorem 12.1.1
(i) If ϕ : G → G0 is a morphism of algebraic groups then ker dϕ =
L(ker ϕ).
(ii) If A, B < G are closed subgroups then a ∩ b = L(A ∩ B).
Proof (i) We may assume that ϕ is surjective. Of course, L(ker ϕ) ⊂

ker dϕ. Since ϕ is separable, dϕ is surjective, and the result follows by
dimensions.
(ii) Let π : G → G/B be the canonical morphism, so ker dπe = b.
Let π 0 : A → π(A) be the restriction of π. The fibers of π 0 are the
cosets of A ∩ B in A, and π 0 is separable. (Also π(A) is a variety
because it is an A-orbit in G/B). Therefore π(A) ∼ = A/(A ∩ B), and
0
now as in (i) we deduce that ker dπe = L(A ∩ B). On the other hand,
ker dπe0 = a ∩ ker dπe = a ∩ b.
Lemma 12.1.2 Let G be connected, ρ : G → GL(V ) be a rational

representation and dρ : g → gl(V ) be the corresponding representation
of g. Then G and g leave the same subspaces (resp. vectors) invariant.
Proof In view of Theorem 12.1.1(i), we may assume that G < GL(V ).

By Problem 9.6.1, L(GL(V )W ) = gl(V )W , and GW = G ∩ GL(V )W ,
gW = g ∩ gl(V )W . By Theorem 12.1.1(ii), L(GW ) = gW . Finally, G
127
128 Characteristic 0 theory
stabilizes W if and only if GW = G and g stabilizes W if and only if
gW = g.
Corollary 12.1.3 Let B be a finite dimensional k-algebra. Then
L(AutB) ∼
= Der B.
Proof The proof of Corollary 9.5.2 shows that x ∈ GL(B) is an automor-

phism if and only if it fixes certain tensor t ∈ B ∗ ⊗B ∗ ⊗B, while X ∈ gl(B)
is a derivation if and only if it kills t. Now apply Lemma 12.1.2.
Definition 12.1.4 Let g = L(G). A Lie subalgebra h of g is called

algebraic if h = L(H) for a closed connected subgroup H < G.
Even in characteristic 0 not all subalgebras are algebraic.
Theorem 12.1.5 Assume that G is connected. Then the map H 7→ h

is a one-to-one inclusion preserving correspondence between the closed
connected subgroups of G and the algebraic Lie subalgebras. Moreover,
normal subgroups correspond to ideals.
Proof Suppose L(H) = L(K). Using Theorem 12.1.1(ii), we have L(H ∩

K) = L(H)∩L(K) = L(H). So dim H∩K = dim H, whence H∩K = H.
Similarly, H ∩ K = K. It follows that H = K.
We already know that h is an ideal if H is normal, see Lemma 9.4.1.
Conversely, suppose h ⊂ g is an ideal. Then g stabilizes h via ad, hence
G stabilizes h via Ad, see Lemma 12.1.2. But for x ∈ G, Ad x : h → g
is the differential of Int x : H → G. By separability, h = Ad x(h) =
L(Int x(H)) = L(xHx−1 ). Now, by the previous paragraph, H =
xHx−1 , as they have the same Lie algebra.
Theorem 12.1.6 Let G be a connected algebraic group.

(i) If x ∈ G, then L(CG (x)) = cg (x).
(ii) ker Ad = Z(G), and L(Z(G)) = z(g).
Proof (i) Lemma 9.4.4 shows that this is true when G = GLn . In
general, embed G as a closed subgroup of some GLn and use Theo-
rem 12.1.1(ii).
(ii) By Theorem 12.1.1, L(ker Ad) = ker ad = z(g). As Ad = d Int,
Z(G) ⊂ ker Ad. Conversely, if x ∈ ker Ad, then g = cg (x) = L(CG (x)),
12.2 Semisimple groups and Lie algebras 129
whence CG (x) = G since they have the same Lie algebras. Thus x ∈
Z(G).
Corollary 12.1.7 A connected algebraic group is commutative if and

only if its Lie algebra is abelian.
12.2 Semisimple groups and Lie algebras

A Lie algebra (of positive dimension) is semisimple if it does not have
non-trivial solvable ideals. This is equivalent to the requirement that
the Lie algebra does not have non-zero commutative ideals. Similarly,
a connected algebraic group of positive dimension is semisimple if and
only if it has no closed connected commutative normal subgroup except
{e}, see Problem 11.3.4.
Theorem 12.2.1 A connected algebraic group is semisimple if and only

if its Lie algebra is semisimple.
Proof If N < G is a closed connected commutative normal subgroup

then n is a commutative ideal of g, so n = 0 forcing N = {e}. Conversely,
let n ⊂ g be a commutative ideal. Define H := CG (n)◦ . Then h = cg (n)
by Lemma 12.1.2. Since n is an ideal, so is cg (n). Hence H is normal
in G. Hence Z := Z(H)◦ is also normal in G. By Theorem 12.1.6(ii),
z is the center of h, and therefore includes n. But G is semisimple, so
Z = {e}, z = 0. This forces n = 0.
Remark 12.2.2 When G is semisimple, g is semisimple, so z(g) = 0,

whence Z(G) is finite, see Theorem 12.1.6.
Corollary 12.2.3 Rational representations of semisimple algebraic groups

are completely reducible.
Proof This follows from the similar fact about Lie algebras (known as
Weyl’s complete reducibility theorem) together with Theorem 12.2.1 and
Lemma 12.1.2.
Theorem 12.2.4 Let G be semisimple. Then Ad G = (Autg)◦ and

ad g = Der g.
130 Characteristic 0 theory
Proof That ad g = Der g is a well-known result in Lie algebras that all
derivations of a semisimple Lie algebra are inner. On the other hand,
Ad G ⊆ Aut(g)◦ , so it suffices to observe that their dimensions coincide.
Well, this follows from dim Ad G = dim G, and dim g = dim Der(g) =
dim Aut(G)◦ , see Corollary 12.1.3.
The theorem shows that a semisimple group can be recovered from
its Lie algebra ”up to a finite center”, and goes a long way towards the
classification of semisimple algebraic groups in characteristic 0.
12.3 Problems
Recall that char k = 0 in this capter.
Problem 12.3.1 Let G be a connected algebraic group, H < G closed

connected subgroup. Prove that L(NG (H)) = ng (h) and L(CG (H)) =
cg (h).
Problem 12.3.2 Let G be a connected algebraic group, h a subalgebra

of g. Prove that L(CG (h)) = cg (h).
Problem 12.3.3 Prove that SL2 is semisimple.

13
Semisimple Lie algebras
We saw that in characteristic 0 a connected algebraic group is semisimple

if and only its Lie algebra is semisimple. Semisimple Lie algebras can
be classified, and this gives us a first approximation to the classification
of semisimple algebraic groups in characteristic 0. It turns out that the
semisimple algebraic group in characteristic 0 is determined up to finite
central subgroup by its Lie algebra (and it is easy to keep the finite
group under control). It turns out that the classification of semisimple
groups is essentially the same in arbitrary characteristic, although this
is much more difficult to prove. In this chapter we are going to review
semisimple Lie algebras and explain how to a semisimple Lie algebra
we can associate an algebraic group in arbitrary characteristic. This is
going to be roughly half of the classification.
13.1 Root systems

We want to review classification of the finite dimensional semisimple Lie
algebras over C. The first step is to introduce the abstract notion of a
root system.
Definition 13.1.1 A root system is a pair (E, Φ) where E is a (real)

Euclidean space and Φ is a finite set of non-zero vectors, called roots, in
E such that
(i) Φ spans E.
(ii) α, cα ∈ Φ implies c = ±1.
(iii) For any root α, Φ is invariant under the reflection sα in the
hyperplane orthogonal to α, i.e. the automorphism
β 7→ β − (β, α∨ )α,
131
132 Semisimple Lie algebras
where α∨ := 2α/(α, α).
(iv) (α, β ∨ ) ∈ Z for all α, β ∈ Z.
Given a root system, the Weyl group W is the subgroup of GL(E)

generated by the sα for α ∈ Φ. It is a finite group, since it acts faithfully
on the finite set Φ.
We let Hα = {β ∈ E | (α, β) = 0} be the hyperplane orthogonal to α.
The connected components of
[
E\ Hα
α∈Φ
are called the Weyl chambers. Fix a chamber C, which we will call the
fundamental chamber. Then one can show that the map
w 7→ wC
is a bijection between W and the set of chambers.

The choice of C fixes several other things. We let Φ+ be the set of
all α ∈ Φ which are in the same half space as C (by this we mean that
(γ, α) > 0 for any γ ∈ C). Then, Φ = Φ+ t (−Φ+ ). Elements of Φ+ are
called positive roots. Next, let
Π = {α ∈ Φ+ | Hα is one of the walls of C}.
This is called a base for the root system. One can show that Π is actually
a basis for the vector space E, and moreover every element of Φ+ is a
non-negative integer linear combination of Π. Elements of Π are called
simple roots.
The Weyl group W is actually generated by the sα for α ∈ Π, i.e. by
the reflections in the walls of the fundamental chamber. This leads to
the idea of the length `(w) of w ∈ W , which is defined as the minimal
length of an expression w = sα1 . . . sαr where α1 , . . . , αr are simple roots.
Geometrically, `(w) is the number of hyperplanes separating wC from
C.
Let Π = {α1 , . . . , α` }. Here ` = dim E is the rank of the root system.
The Cartan matrix A = (ai,j )1≤i,j≤` is the matrix with
ai,j = (αi , αj∨ ).
Since all the Weyl chambers are conjugate under the action of W , the
Cartan matrix is an invariant of the root system (up to simultaneous
permutation of rows/columns). Here are some basic properties about
this matrix:
13.1 Root systems 133
(C1) ai,i = 2.
(C2) For i 6= j, ai,j ∈ {0, −1, −2, −3}.
(C3) ai,j 6= 0 if and only if aj,i 6= 0.
Note (C2) is not obvious. It follows because E 0 = hαi , αj i together with

Φ0 := Φ ∩ E 0 is a root system of rank 2. Rank 2 root systems are easy
(and fun) to classify. Their Cartan matrices are exactly the following:

2 0 2 −1
A1 × A1 : , A2 : ,
0 2 −1 2

2 −2 2 −1
B2 : , G2 : .
−1 2 −3 2
Note if ai,j 6= 0, then
(αi , αi )/(αj , αj ) = ai,j /aj,i ∈ {1, 2, 3},
so in this case you can work out the ratio of the lengths of the roots
αi , αj to each other from the Cartan matrix.
A root system is called indecomposable if it cannot be partitioned
E = E1 ⊥ E2 , Φ = Φ1 t Φ2 where (Ei , Φi ) are root systems. An equiv-
alent property is that we cannot order roots in such a way that the
corresponding Cartan matrix has block-diagonal form. Thus, for an in-
decomposable root system, one can work out the ratio of lengths of any
pair of roots to each other from the Cartan matrix, hence one completely
recovers the form (., .) on E up to a scalar from the Cartan matrix. One
also recovers Φ, since the Cartan matrix contains enough information to
compute the reflection sαi for each i = 1, . . . , `, and Φ = W Π. So (with
the correct definition of an isomorphism—give it!) an indecomposable
root system is completely determined up to isomorphism by its Cartan
matrix.
A convenient shorthand for Cartan matrices is given by the Dynkin
diagram. This is a graph with vertices labelled by α1 , . . . , α` . There are
ai,j aj,i edges joining vertices αi and αj , with an arrow pointing towards
αi if (αi , αi ) < (αj , αj ) Clearly you can recover the Cartan matrix from
the Dynkin diagram given properties (C1)–(C3) above.
Now I can state the classification of root systems:
Theorem 13.1.2 The Dynkin diagrams of the indecomposable root sys-

tems are as given in Figure 13.1.
An • • • •... • • • •
Bn • < • • •... • • • •
Cn • > • • •... • • • •
•H
Dn •
H • •... • • • •
•
E6 • • • • •
•
E7 • • • • • •
•
E8 • • • • • • •
•
F4 • • < • •
G2 • < •
Fig. 13.1. Dynkin diagrams of semisimple Lie algebras
13.2 Semisimple Lie algebras

Now we sketch how the semisimple Lie algebras are classified by the root
systems. We need to start with a semisimple Lie algebra and build a
root system out of it, and vice versa.
So we begin with a finite dimensional semisimple Lie algebra g over
C. Then g possesses a non-degenerate invariant symmetric bilinear form
(., .), where invariant here means ([X, Y ], Z) = (X, [Y, Z]) (in fact, the
converse is also true). Moreover, if g is simple, there is a unique such
form up to a scalar. There is a “canonical” choice of non-degenerate
form, the Killing form, but we don’t need that here.
Example 13.2.1 Let us consider sln . The bilinear form (X, Y ) =

tr(XY ) is non-degenerate and invariant. Let ei,j be the ij-matrix unit
and let h be the diagonal, trace zero matrices. We can decompose
M
sln = h ⊕ Cei,j .
i6=j
A basis for h is given by h1 , . . . , hn−1 where hi = ei,i − ei+1,i+1 . Let

13.2 Semisimple Lie algebras 135
εi ∈ h∗ be the map sending a diagonal matrix to its ith diagonal entry.
Note ε1 + · · · + εn = 0, i.e. the εi ’s are not independent. Then for any
H ∈ h we have
[H, ei,j ] = (εi − εj )(H)ei,j ,
i.e. ei,j is a simultaneous eigenvector for h. We use the word weight in
place of eigenvalue, so ei,j is a vector of weight εi − εj . Now you recall
that the root system of type An−1 can be defined as the real vector
subspace of h∗ spanned by ε1 , . . . , εn , and the roots are
Φ := {εi − εj | i 6= j}.
A base for Φ is given by by α1 , . . . , αn−1 where αi = εi − εi+1 . Let us
finally write gα := Cei,j if α = εi − εj , i.e. the weight space of g of
weight εi − εj . Then
M
g=h⊕ gα .
α∈Φ
In other words, you “see” the root system of type An−1 when you decom-
pose g into weight spaces with respect to the diagonal matrices. Final
note: the inner product giving the Euclidean space structure is induced
by the non-degenerate form defined to start with. Indeed if you compute
the matrix (hi , hj ) you get back the Cartan matrix of type An−1 .
This example is more or less how things go in general, when you

start with an arbitrary semisimple Lie algebra g, with a non-degenerate
invariant form (., .). The first step is to develop in g a theory of Jordan
decompositions. This parallels the Jordan decomposition we proved for
algebraic groups. You call an element X of g semisimple if the linear
map ad X : g → g is diagonalizable, and nilpotent if ad X is nilpotent.
The abstract Jordan decomposition shows that any X ∈ g decomposes
uniquely as X = Xs + Xn where Xs ∈ g is semisimple and Xn ∈ g is
nilpotent, and [Xs , Xn ] = 0.
What is more, if you have a representation of g, i.e. a Lie algebra
homomorphism ρ : g → gln , it is true that ρ(Xs ) = ρ(X)s and ρ(Xn ) =
ρ(X)n , where the semisimple and nilpotent parts on the right hand
side are taken just as n × n matrices in gln . Thus, the abstract Jordan
decomposition is consistent with all other Jordan decompositions arising
from all other representations. In particular, semisimple elements of g
map to diagonalizable matrices under any matrix representation of g.
For sln , ei,j is nilpotent for i 6= j, and ei,i − ei,j is semisimple.
Now you introduce the notion of a maximal toral subalgebra or Cartan
subalgebra h of g (in general maximal toral subalgebra and Cartan subal-
gebra are different notions but they agree for semisimple algebras). This
is a maximal abelian subalgebra all of whose elements are semisimple.
It turns out that in a semisimple Lie algebra, maximal toral subalge-
bras are non-zero, and they are all conjugate under automorphisms of g.
Now fix one – it doesn’t really matter which, since they are all conjugate.
Importantly, the restriction of the invariant form (., .) on g to h is still
non-degenerate. So we can define a map
h∗ → h
mapping α ∈ h∗ to tα ∈ h, where tα is the unique element satisfying

(tα , h) = α(h) for all h ∈ h. Now we can lift the non-degenerate form on
h to h∗ by defining (α, β) = (tα , tβ ). Thus, h∗ now has a non-degenerate
symmetric bilinear form on it too.
For α ∈ h∗ , define
gα = {X ∈ g | [H, X] = α(H)X for all H ∈ g}.
6 α ∈ h∗ | gα 6= 0}. Then you get

L
Clearly, g = α∈h∗ gα . Set Φ = {0 =
Cartan decomposition of g:
M
g=h⊕ gα
α∈Φ
(it is not obvious that the right hand side is everything...). It turns out
with some work that each of the gα spaces are one-dimensional.
Now you can build a root system out of g: we’ve already constructed
the set Φ. Let E be the real vector subspace of h∗ spanned by Φ. The
restriction of the form on h∗ to E turns out to be real valued only, and
makes E into a Euclidean space. Now:
Theorem 13.2.2 The pair (E, Φ) just built out of g (starting from a
choice of h) is a root system. Moreover, the resulting map from semisim-
ple Lie algebras to Dynkin diagrams gives a bijection between isomor-
phism classes of semisimple Lie algebras and Dynkin diagrams. The
decomposition of a semisimple Lie algebra as a direct sum of simples
corresponds to the decomposition of the Dynkin diagram into indecom-
posable components.
For example, sln is the simple Lie algebra corresponding to the Dynkin
diagram An−1 .
13.3 Construction of simple Lie algebras 137
13.3 Construction of simple Lie algebras
We now explain how to construct the simply-laced simple Lie algebras.
So let (E, Φ) be a root system of type A` , D` or E` , π = {α1 , . . . , α` } be
a base and Φ+ the corresponding set of positive roots. We may assume
that (α, α) = 2 for all α ∈ Φ, as all roots have the same length. Let
Q = ZΦ ⊂ E be the root lattice, the free abelian group on basis Π.
We construct an asymmetry function
ε : Q × Q → {±1}
such that
(1) ε is bilinear, i.e. ε(α + α0 , β) = ε(α, β)ε(α0 , β) and ε(α, β + β 0 ) =

ε(α, β)ε(α, β 0 ) for all α, α0 , β, β 0 ∈ Q.
(2) ε(α, α) = (−1)(α,α)/2 for all α ∈ Q.
Note (2) implies
(3) ε(α, β)ε(β, α) = (−1)(α,β) for all α, β ∈ Q.
To construct such an ε, it suffices by bilinearity to define it on elements

of Π. Choose an orientation of the Dynkin diagram. Then define


 1 if αi and αj are not connected,
1 if αi → αj ,

ε(αi , αj ) =

 −1 if αi ← αj ,

−1 if αi = αj .
Now we can construct g. Let h∗ = C ⊗Z Q = C ⊗R E. Let h be the

dual space, and let Hα ∈ h be the element such that β(Hα ) = (β, α) for
all β ∈ h∗ . Then, H1 , . . . , H` gives a basis for h, where Hi = Hαi .
Now let
M
g=h⊕ CEα
α∈Φ
as a vector space. Define a multiplication by the formulae
[Hi , Hj ] = 0,
[Hi , Eα ] = α(Hi )Eα = (αi , α)Eα ,
[Eα , E−α ] = −Hα ,
[Eα , Eβ ] = 0 if α + β ∈
/ ϕ ∪ {0},
[Eα , Eβ ] = ε(α, β)Eα+β if α + β ∈ Φ.
Theorem 13.3.1 g is the simple Lie algebra of type Φ, with maximal
toral subalgebra h.
Proof You of course have to check that g is a Lie algebra, which boils
down to checking that the Jacobi identity is satisfied. This is a case
analysis.
Having done that, we define a bilinear form on g by
(Hi , Hj ) = (αi , αj ), (Hi , Eα ) = 0, (Eα , Eβ ) = −δα,−β .
You check that this is a non-degenerate invariant bilinear form. More-

over, h is a toral subalgebra of g, and since the 0-weight space of h on
g is just h itself, it must be maximal. Finally, it is automatic that the
corresponding root system is of type Φ. Hence, g is simple of type Φ
with maximal toral subalgebra h.
Definition 13.3.2 Let g be an arbitrary semisimple Lie algebra (not

necessarily simply-laced). Let Φ be a root system corresponding to a
choice of maximal toral subalgebra h, and let Π = {α1 , . . . , α` } be a
base for Φ. For α, β ∈ Φ, the α-string through β is the sequence
β − rα, . . . , β, . . . , β + qα
where r and s are the maximal integers such that all the vectors in the
string belong to Φ. It turns out that r and q are equal to 0, 1, 2 or 3 in
all cases, and 2 and 3 don’t arise if the root system is simply-laced.
Denote Hα := 2tα /(α, α) and Hi := Hαi . A Chevalley basis for g
means a basis
{H1 , . . . , H` } ∪ {Xα | α ∈ Φ}
such that
(a) [Hi , Hj ] = 0,
(b) [Hi , Xα ] = (α, αi∨ )Xα ,
(c) [Xα , X−α ] = Hα , and this is a Z-linear combination of H1 , . . . , H` ,
(d) If α, β, α + β ∈ Φ and β − rα, . . . , β + qα is the α-string through
β, then [Xα , Xβ ] = Nα,β Xα+β = ±(r + 1)Xα+β .
The key thing is that all the structure constants in a Chevalley basis are
integers!
Theorem 13.3.3 (Chevalley) Chevalley bases exist.

13.4 Kostant Z-form 139
Proof If Φ is simply-laced, this is easy from the above construction:
take Xα = Eα if α ∈ Φ+ and −Eα is α ∈ Φ− . Now you easily check
this satisfies the properties. If Φ is not simply-laced, we need some
other construction. For classical Lie algebras that is not too hard: you
can write them down just as explicitly as sln . Problem 13.6.4 gives an
example of how you do this. Another way is to realize all the non-simply-
laced root systems as fixed points of automorphisms of simple-laced ones.
13.4 Kostant Z-form

Informally speaking, Chevalley group is constructed from a semisimple
Lie algebra g as the group generated by the ‘exponents’ of the form
exp(tXα ) where Xα is a root element of the Lie algebra and t is a scalar.
But there are some problems here. Consider, for example
exp(Xα ) = 1 + Xα + Xα2 /2! + Xα3 /3! + . . .
What does that mean? We don’t have a toplogy to speak of convergence,

so we need to make sure that the sum is finite. Well, this will be achieved
if Xα is nilpotent in a certain sense. Further, what does Xα3 mean? We
can’t multiply in a Lie algebra! However we can consider this as an
element of the universal enveloping algebra. There is a further problem
however. If characteristic is 2 or 3, we can’t make sense of the division
by 3!. The solution to this is very clever—we will first divide by 3! and
then pass to characteristic p!! More formally, we will consider a Z-form
UZ of the universal enveloping algebra U of g which contains all Xαn /n!
and then pass to the algebra U = Uk := UZ ⊗Z k, called the hyperalgebra.
First, recall the universal enveloping algebra U (g) associated to a Lie
algebra g. It is defined by a universal property, but there is also an
explicit construction. The all-important PBW theorem shows that we
can identify g with a Lie subalgebra of U (g), and moreover if X1 , . . . , XN
is a basis for g, then the monomials
X1r1 . . . XN
rn
give a basis for U (g).

One reason U (g) is important is because it allows you to view rep-
resentations, i.e. Lie algebra homomorphisms ρ : g → gl(V ), as U (g)-
modules:
Lemma 13.4.1 The categories of representations of g and of U (g)-
modules are isomorphic.
Proof Let ρ : g → gl(V ) be a representation. Viewing gl(V ) as End (V ),

we get induced a unique associative algebra homomorphism ρ̂ : U (g) →
End (V ). Using this, we make V into a U (g)-module by u.v = ρ̂(u)(v). If
you think about it, this gives a functor {representations of g} → {U (g)-
modules}. Conversely, given a U (g)-module, define ρ : g → gl(V ) by
ρ(X)(v) := Xv. This defines an inverse functor.
Now we state the main result about the Kostant Z-form.
Theorem 13.4.2 Let g be a semisimple Lie algebra over C, with Cheval-

ley basis {H1 , . . . , H` } ∪ {Xα | α ∈ Φ}. Let UZ be the Z-subalgebra of
U (g) generated by the Xαr /r! for all α ∈ Π, r ≥ 1. Then, UZ is free as a
Z-module with basis given by all monomials in the

Hi
Xαrα /|rα ! (α ∈ Φ), (i = 1, . . . , `)
mi
in some fixed order, where mi , rα ≥ 0.
Proof (1) Observe all the “Kostant monomials” form a C-basis for U (g)
by the PBW theorem, so they are linearly independent.

(r) Hi
(2) Observe all Xα and all belong to UZ – by constructing
mi
them as various commutators of the generators of UZ . Hence all Kostant
monomials belong to UZ .
(3) Prove that the product of two Kostant monomials can be expanded
as a Z-linear combination of other Kostant monomials. Hence they span
UZ . This is done by proving various commutation relations.
13.5 Weights and representations

∗
Let Q = ZΦ ⊂ h be the root lattice. Let P be the weight lattice, defined
as
P = {λ ∈ h∗ | λ(Hi ) ∈ Z for all i = 1, . . . , `}.
Thus P is the lattice dual to the lattice ZH1 +· · ·+ZH` in h. Obviously,

Q ⊆ P . Moreover, since P and Q are both lattices in h∗ , i.e. they are
both finitely generated abelian groups that span h∗ over C, the quotient
13.5 Weights and representations 141
P/Q is a finitely generated, torsion abelian group. But that implies P/Q
is a finite abelian group. It is called the fundamental group.
To get a basis for P (as a free abelian group), one can take the fun-
damental weights ω1 , . . . , ω` defined by
ωi (Hj ) = δi,j ,
i.e. the dual basis to H1 , . . . , H` . We claim that
`
X
αi = ai,j ωj
j=1
where ai,j = (αi , αj∨ ) = αi (Hj ) is the Cartan integer. To see this, just
evaluate both sides on Hj – you get the same thing. Thus, P/Q is the
abelian group on generators ω̄1 , . . . , ω̄j subject to relations
`
X
ai,j ω̄j = 0.
j=1
Considering elementary divisors, you get that

|P/Q| = det A,
indeed, you get an explicit description of P/Q as an abelian group.
These are the orders:
A` : ` + 1
B ` , C` , E 7 : 2
D` : 4
E6 : 3
E8 , F 4 , G 2 : 1
In fact the fundamental group is cyclic in all cases except for D` with `
even, when it is Z/2 × Z/2. This is going to be very important: we now
know exactly all possible lattices lying between Q and P .
The last important ingredient that we need to construct the Chevalley
groups is a little representation theory of semisimple Lie algebras. We
are interested here just in the finite dimensional representations of g. A
fundamental theorem of Weyl (mentioned before) shows that any finite
dimensional representation of g decomposes as a direct sum of irrducible
representations, i.e. ones with no proper invariant submodules. So we
really only need to discuss the finite dimensional irreducible representa-
tions.
Now, if V is a finite dimensional U (g)-module, we can decompose
M
V = Vλ
λ∈h∗
where
Vλ = {v ∈ V | Hv = λ(H)v for all H ∈ h}.
This is the weight space decomposition of V . For example, the Cartan

decomposition of g itself is the weight space decomposition of the adjoint
representation. We will say that v ∈ V is a high weight vector of high
weight λ if 0 6= v ∈ Vλ and Xα v = 0 for all α ∈ Φ+ . It is obvious that
any non-zero finite dimensional V possesses such a vector – because
Xα Vλ ⊆ Vλ+α and there are only finitely many non-zero weight spaces
in total.
It turns out that the high weight vector of an irreducible representa-
tion of g is unique up to a scalar and its weight belongs to the set
P + = {λ ∈ P | λ(Hi ) ≥ 0 for each i = 1, . . . , `}.
Conversely, for any element of P + , there is a unique up to isomorphism

irreducible representation of that highest weight.
Relation to Z-forms is as follows:
Theorem 13.5.1 Let λ ∈ P + and V be the corresponding irreducible

highest weight representation. Then, there exists a lattice VZ in V in-
variant under the action of the Kostant Z-form UZ , such that
X
VZ = Vµ,Z
µ∈h∗
where Vµ,Z = Vµ ∩ VZ .
Given any finite dimensional representation V , we consider its lattice

L(V ), which is defined to be the subgroup of P generated by all λ such
that Vλ 6= 0. If V is faithful, then Q ⊆ L(V ) ⊆ P . In fact you can get
any intermediate lattice arising for suitable choice of V , and the possible
lattices are parametrized by the subgroups of the fundamental group.
13.6 Problems
Problem 13.6.1 Write down the explicit construction of the root sys-
tems of type A` , B` , C` and D` , and show that the length of the longest
13.6 Problems 143
element w0 of the Weyl group was `(` + 1)/2. (Hint: You need to look it
up! There are many good sources, e.g. Humphreys’ “Introduction to Lie
algebras and representation theory”, Bourbaki “Groupes et Algebres de
Lie”, Kac “Infinite dimensional Lie algebras”, Carter “Finite groups of
Lie type”, Helgason “Differential geometry and symmetric spaces”...)
Problem 13.6.2 Look up or work out the dimensions of the simple Lie
algebras of types A` , B` , C` and D` . In particular, check that dim C` is
the same as the dimension of the algebraic group Sp2` .
Problem 13.6.3 In the proof of Theorem 13.3.1, go through the details

needed to verify that the bilinear form defined is invariant.
Problem 13.6.4 Let V be a (2` + 1)-dimensional complex vector space

with an ordered basis e1 , . . . , e` , e0 , e−` , . . . , e−1 . Define a symmetric
bilinear form on V by declaring (ei , ej ) = 0 (i 6= −j), (ei , e−i ) = 1
(i 6= 0) and (e0 , e0 ) = 2. Let J be the matrix of this bilinear form in the
basis, ordering rows and columns as e1 , . . . , e` , e0 , e−` , . . . , e−1 .
(i) Compute the matrix J explicitly.
(ii) Let g = {X ∈ gl(V ) | X T J + JX = 0} be the Lie algebra so(V ) =
so(2` + 1). Viewing elements of g as block matrices in our ordered basis,
we can write
 
A v B
X =  r x s .
 
C w D
Compute explicitly the conditions that the ` × ` matrices A, B, C, D, the

column vectors v, w, the row vectors r, s and the scalar x must satisfy
for X to belong to g.
(iii) Let h be the set of all diagonal matrices in g. This is a toral
subalgebra of g. Let i ∈ h∗ be the function
diag(t1 , . . . , t` , 0, −t` , . . . , −t1 ) 7→ ti ,
so that ε1 , . . . , ε` form a basis for h∗ . Let
Φ = {±εi ± εj , ±εk | 1 ≤ i < j ≤ `, 1 ≤ k ≤ `},
the root system of type B` . Let Ei,j : V → V denote the linear map
with Ei,j .vk = δjk vi for all −` ≤ i, j, k ≤ `. For α ∈ Φ, define
α i − j (i < j) i + j (i < j) i
Xα Ei,j − E−j,−i Ej,−i − Ei,−j 2Ei,0 − E0,−i
X−α Ej,i − E−i,−j E−i,j − E−j,i E0,i − 2E−i,0
Verify that
M
g=h⊕ CXα
α∈Φ
is the Cartan decomposition of g.

(v) Defining H1 , . . . , H` appropriately, check that {H1 , . . . , H` }∪{Xα |α ∈
Φ} is a Chevalley basis for g.
14
The Chevalley construction
To motivate the construction, let’s stick to working over C for a bit. Let
g be a semisimple Lie algebra, and let V be a finite dimensional faithful
representation. So the lattice L(V ) is some intermediate lattice between
Q and P (e.g. if V is the adjoint representation, L(V ) = Q).
Let {Hi } ∪ {Xα } be a Chevalley basis. Since V has only finitely
many weight spaces and the Xα map Vµ into Vµ+α , each Xα acts on V
nilpotently. Thus, we can consider the formal series
exp(cXα ) = 1 + cXα + c2 Xα2 /2! + . . .
for any scalar c ∈ C as a well-defined endomorphism of V (the infinite
sum terminates...).
By familiar properties of exponential series,
exp(cXα ) exp(dXα ) = exp((c + d)Xα ).
In particular, exp(cXα ) is invertible with inverse exp(−cXα ). Now let
G be the subgroup of GL(V ) generated by all exp(cXα ) for all c ∈ C
and α ∈ Φ. This is the Chevalley group corresponding to g in the
representation V . It turns out that (up to isomorphism) G is determined
by g and the lattice L(V ).
Using Z-forms we can imitate this construction over an arbitrary field
k. In the case that k is an algebraically closed field, G is always a
semisimple algebraic group, and in fact all semisimple algebraic groups
arise out of this contstruction for some choice of Φ and L(V ).
In order to study the structure of G in detail, we construct closed
subgroups U, T, B, N of G with explicitly named generators, and prove
various relations between these generators. We will show that B = U oT
(semidirect product), that T N , and identify the quotient group N/T
with the original Weyl group W .
145
146 The Chevalley construction
Finally, we will prove the Bruhat decomposition:
[
G= BwB.
w∈W
Here w ∈ W needs to be interpreted as a fixed coset representative in

N . Thus, the (B, B)-double cosets in G are parametrized by the Weyl
group W .
14.1 Definition and first properties

To every field, every root system Φ and every lattice L such that Q ⊆
L ⊆ P we associate the Chevalley group G = G(k, Φ, L) defined as fol-
lows. Let UZ be the Kostant Z-form of the universal enveloping algebra
U (g) of the semisimple Lie algebra g of type Φ, and VC be a represen-
tation of g with L = L(V ). Pick a UZ -invariant lattice VZ in V as in
Theorem 13.5.1. Let V = VZ ⊗Z k. Then for any t ∈ k and α ∈ Φ,
xα (t) := exp(tXα ) = 1 + Xα ⊗ t + (Xα2 /2!) ⊗ t2 + . . .
can be considered as an invertible endomorphism of V . By definition, G
is the group generated by all xα (t) for t ∈ k and α ∈ Φ.
The proof that the group only depends on L and not on the choice of
VC and VZ ⊂ VC will be skipped.
For fixed α ∈ Φ, let Xα be the subgroup of GL(V ) generated by all
xα (t) for all t ∈ k.
From now on we assume as usual that k is algebraically closed.
Theorem 14.1.1 The group G is a closed connected subgroup of GL(V ).
Proof Note that the map Ga → GL(V ), t 7→ xα (t) is a morphism of

algebraic groups, as the exponent stops after finitely many steps and
so is ”polynomial” in t. So each Xα is a closed connected subgroup of
GL(V ). Now use Corollary 8.2.7.
The main goal of this course is to prove that: (a) G is a semisimple
algebraic group and (b) every semisimple algebraic group is obtained in
this way. You have probably already figured out that we are not going
to get there by the end of the term but we will try to get as far as
possible...
Concerning (a), it is a well known fact (often proved even in the 600
algebra courses) that P SLn (k) is simple as an abstract group (all you
need for this is the assumption that k has more than 3 elements if n = 2,
14.1 Definition and first properties 147
but remember that for us k is algebraically closed, so infinite). So if G
is of type A, any of its solvable normal subgroups is contained in the
center of G, which is finite. As the radical is connected, this proves that
the radical of G is trivial. The proof for other types uses elements xα (t)
instead of transvections in the usual proof for P SLn , and we will not
reproduce it here.
15
Borel subgroups and flag varieties
In the previous chapter, we sketched the construction of the semisimple

algebraic groups. It is very explicit and case-free. We will now go back to
algebraic geometry and sketch the proof of the Classification Theorem.
We will see that algebraic geometry, in particular the variety structure
on quotient varieties G/H which we haven’t really used yet in a deep
way, is a fundamental tool to studying group theory.
15.1 Complete varieties and Borel’s fixed point theorem

Recall the notion of the complete variety from chapter 7. We need the
following:
Lemma 15.1.1 Let G be an algebraic group acting transitively on va-

rieties X, Y . Let ϕ : X → Y be a bijective, G-equivariant morphism. If
Y is complete, then X is complete.
Proof By Remark 7.1.3(ii), we need to show that π2 : X × Z → Z

is closed for all affine varieties Z. Since Y is complete, it suffices to
prove that ϕ × id : X × Z → Y × Z is closed. By Proposition 5.3.2,
there are open subsets U ⊂ X and V ⊂ Y such that ϕ(U ) = V and
ϕ|U : U → V is a finite morphism. Let R, S, T be the respective affine
algebras of U, V, Z. Since R is integral over S, R ⊗ T is integral over
S ⊗ T . Therefore ϕ × id : U × Z → V × Z is also a finite morphism.
In particular, it is a closed map, see Corollary 5.2.4. Because G acts
transitively on X, Y and ϕ is G-equivariant, X (resp. Y ) is covered by
finitely many translates of the form xU (resp. xV ) for some x ∈ G. It
follows that ϕ × id : X × Z → Y × Z is closed.
148
15.2 Borel subgroups 149
Now we can prove the important
Theorem 15.1.2 (Borel’s fixed point theorem) Let G be a con-

nected solvable algebraic group, and X be a non-empty complete G-
variety. Then G has a fixed point on X.
Proof Proceed by induction on dim G, the case G = {1} being trivial.

Suppose then that dim G > 0 and let H = G0 , which is connected
solvable of strictly smaller dimension. By induction,
Y = {x ∈ X | Hx = x}
is non-empty. By Lemma 8.3.1(iii), Y is closed, hence complete, and G
stabilizes Y as H G. So we may as well replace X by Y to assume
that H ⊆ Gx for all x ∈ X. Since G/H is abelian, this implies that each
Gx G.
Now choose x so that the orbit G · x is of minimal dimension. Then,
G · x is closed hence complete. The map G/Gx → G.x is bijective, so
we deduce that G/Gx is complete by the preceeding lemma. But G/Gx
is affine as Gx G. So in fact G/Gx is a point, i.e. G = Gx and x is a
fixed point.
Corollary 15.1.3 (Lie-Kolchin theorem) Let G be a connected solv-

able subgroup of GL(V ). Then G fixes a flag in V .
Proof Let G act on the flag variety F(V ). This is projective, so G has
a fixed point.
15.2 Borel subgroups

Let G be a connected algebraic group.
Definition 15.2.1 A Borel subgroup B of G is a maximal closed con-

nected solvable subgroup of G.
Example 15.2.2 (i) If G is a Chevalley group, the subgroup B = T U is

a Borel subgroup of G. Any conjugate of B in G will give another such
subgroup.
(ii) If G = GLn , the subgroup B of all upper triangular matrices
is a maximal closed connected solvable subgroup by the Lie-Kolchin
theorem. Hence, it is a Borel subgroup. Any conjugate of this will give
150 Borel subgroups and flag varieties
another. Note in this case that the quotient variety G/B is the flag
variety, so it is a projective variety in particular.
Theorem 15.2.3 For any connected algebraic group G, let B be a Borel

subgroup. Then, G/B is a projective variety, and all other Borel sub-
groups of G are conjugate to B.
Proof Let S be a a Borel subgroup of maximal dimension. Apply Cheval-

ley’s theorem to construct a representation ρ : G → GL(V ) and a 1-space
L ⊂ V such that S = StabG (L). By the Lie-Kolchin theorem, S fixes a
flag in V /L. Hence S fixes a flag F = (L = L1 ⊂ L2 ⊂ · · · ⊂ Ln = V )
in the flag variety F(V ). Recall this is a projective variety, hence it is
complete.
By the choice of L, S = StabG (F ). Hence the orbit map induces a
bijective morphism G/S → G · F ⊂ F(V ). Take any other flag F 0 ∈
F(V ). Then StabG (F 0 ) is upper triangular in some basis, hence it is
solvable, hence its dimension is ≤ dim S. This shows that dim G · F 0 =
dim G − dim StabG (F 0 ) ≥ dim G · F . Therefore, G · F is a G-orbit in
F(V ) of minimal dimension, hence it is closed. This shows that G · F
is also complete, so G/S is complete too. Now G/S is complete and its
quasi-projective, hence it is projective.
Finally, let B be another Borel subgroup of G. Then B acts on G/S,
so by Borel’s fixed point theorem, B has a fixed point gS on G/S.
Therefore BgS = gS, i.e. g −1 Bg ⊆ S. By maximality, we get that
g −1 Bg = S and this completes the proof.
Definition 15.2.4 A parabolic subgroup P of G is any closed subgroup

of G such that G/P is a projective (equivalently, complete) variety.
Theorem 15.2.5 Let P be a closed subgroup of G. Then, P is parabolic

if and only if it contains a Borel subgroup. In particular, P is a Borel
subgroup if and only if P is connected solvable and G/P is a projective
variety.
Proof Suppose G/P is projective. Let B be a Borel subgroup of G.

It acts on G/P with a fixed point, say BgP = gP . This implies that
g −1 Bg ⊆ P , i.e. P contains a Borel subgroup.
Conversely, suppose P contains a Borel subgroup B. The map G/B →
G/P is surjective and G/B is complete. Hence, G/P is complete too.
But it is quasi-projective too, so in fact G/P is projective.
Example 15.2.6 (i) Let G = GLn . The subgroups of G containing B
(upper triangular matrices) are exactly the “step” subgroups, one for
each way of writing n = n1 + · · · + ns as a sum of positive integers
n1 , . . . , ns . There are 2n−1 such subgroups.
(ii) Let G be an arbitrary Chevalley group. Let S be a subset of
the simple roots Π, so there are 2` possibilities for S. Let P be the
subgroup of G generated by B and all sα for α ∈ S. (Equivalently, P is
the subgroup generated by T and the Xα ’s for α ∈ Φ+ ∪ (−S).) Then,
P contains B so it is a parabolic subgroup by the theorem, and G/P is
a projective variety. In fact, these P ’s give all the parabolic subgroups
of G containing the fixed choice of Borel subgroup B. By the theorem,
all other parabolic subgroups of G are conjugate to one of these. There
are exactly 2` different conjugacy classes of parabolic subgroup in the
Chevalley group G.
15.3 The Bruhat order

Let G be a Chevalley group, with all the subgroups U, T, Xα , B, N, W =
N/T ,. . . . Recall also that W is generated by the simple reflections
{sα | α ∈ Π}. For any w ∈ W , we can write w as a product of simple
reflections. The length of w was the length of a shortest such expression,
called a reduced expression for W .
Let me define a partial order on W as follows. Take w, w0 ∈ W . Let
w = s1 . . . sr be a reduced expression for w. Declare that w0 ≤ w if
and only if w0 = si1 . . . sij for some 1 ≤ i1 < · · · < ij ≤ r, i.e. if w0
is a “subexpression” of w. How do you prove this really is a partial
order? How do you even show that it is well-defined, i.e. independent
of the choice of the reduced expression of w? One of the ways is to use
algebraic geometry!
By the Bruhat decomposition, the B-orbits on G/B are parametrized
by the Weyl group W , i.e. the orbits are the BwB/B’s. Now, the closure
of an orbit is a union of orbits, the ones in the boundary being of strictly
smaller dimension. So there is obviously a partial ordering ≤ on the
orbits of B on G/B defined by O ≤ O0 if and only if O ⊆ O0 . What
we are going to prove is that this is exactly the partial ordering defined
in the previous paragraph! In other words, the ordering in the previous
paragraph IS well-defined because there is a geometrically defined partial
ordering that amounts to the combinatorics there.
Let’s proceed with some lemmas.
152 Borel subgroups and flag varieties
Lemma 15.3.1 Let G be an algebraic group, X a G-variety, H ≤ G a
closed subgroup and Y ⊆ X a closed H-stable subvariety of X. If G/H
is a complete variety (i.e. if H is a parabolic subgroup of G) then G · Y
is closed in X.
Proof Let A = {(g, x) ∈ G × X | g −1 x ∈ Y }, which is closed in G × X.

Let
π : G × X → G/H × X
be the quotient map. Recall this is an open map. If (g, x) ∈ A then

(gh, x) ∈ A for all h ∈ H, since H stabilizes Y . Hence,
π(A) = G/H × X − π(G × X − A),
so π(A) is closed in G/H × X. Since G/H is complete, the projection

prX (π(A)) ⊆ X is also closed. This is exactly G · Y .
Lemma 15.3.2 Any product of parabolic subgroups of G containing B

is closed in G.
Proof Let P1 , . . . , Pr be parabolic subgroups of G containing B. By

induction, P2 . . . Pr is closed in G and B-stable. Note P2 . . . Pr .B =
P1 . . . Pr−1 is closed and B-stable. Since P1 /B is complete, we get by
the lemma that P1 P2 . . . Pr is closed too.
Theorem 15.3.3 (Chevalley) Let G be a Chevalley group. Fix w ∈ W

and a reduced expression w = s1 . . . sr for w as a product of simple
reflections. Then,
[
BwB = Bw0 B
w0
where w0 runs over all subexpressions si1 . . . sij of s1 . . . sr .
Proof Let w = s1 . . . sr be the fixed reduced expression for w, Let
Pi = hB, si i = B ∪ Bsi B,
where the last equality comes from the work on Chevalley groups in the
previous chapter. We show by induction on r that
[
P1 . . . Pr = Bw0 B
w0
where the union is taken over all subexpressions w0 of the reduced ex-
pression s1 . . . sr . The case r = 1 is already done!
Now suppose r > 1. Then by induction,
[
P1 . . . Pr = Bw00 B.(B ∪ Bsr B)
w00
00
where w runs over all subexpressions of s1 . . . sr−1 . But that equals
[
Bw 0 B
w0
as required, since Bw sr B ⊆ Bw00 BBsr B ⊆ Bw00 B ∪ Bw00 sr B by the

00
previous chapter.
By the preceeding lemma, P1 . . . Pr is closed, hence
[
Bw0 B,
w0
0
union over all subexpressions w of s1 . . . sr , is closed. So it certainly con-
tains the closure BwB. Finally, we know dim BwB is equal to dim B+
the number of positive roots sent to negative roots by w. So in fact we
must have that
[
Bw0 B = BwB
w0
by dimension.
Now since BwB is defined intrinsically independent of any choice of
reduced expression of w, the relation w0 ≤ w iff w0 is a subexpression
of some fixed reduced expression for w is well-defined independent of
the choice. Moreover, it is a partial ordering on W called the Bruhat
ordering.
For w ∈ W , the Schubert variety
Xw := BwB/B
is a closed subvariety of the flag variety G/B. Note Xw is no longer an
orbit of an algebraic group, so it needn’t be smooth. Schubert varieties
are extremely interesting projective varieties with many wonderful prop-
erties. The Schubert variety Xw0 is the flag variety itself, the Schubert
variety X1 is a point. We have shown above that in general, the lattice
of containments of Schubert varieties is isomorphic to the Bruhat order
on W .
16
The classification of reductive algebraic
groups
16.1 Maximal tori and the root system

Now we sketch the procedure to build a root system starting from an
arbitrary reductive algebraic group. This is the first step in proving the
classiciation of reductive algebraic groups.
Let us start by talking about tori. Recall an n-dimensional torus is an
algebraic group isomorphic to Gm ×· · ·×Gm . For example, the subgroup
Dn of GLn consisting of all diagonal matrices is an n dimensional torus.
Let T be an n-dimensional torus. The character group
X(T ) = Hom(T, Gm ) ∼
= Hom(Gm , Gm )⊕n ∼
= Zn .
An important point is that, given any two tori T and T 0 ,
Hom(T, T 0 ) ∼
= Hom(X(T 0 ), X(T )).
So any homomorphism f : X(T 0 ) → X(T ) of abelian groups induces a

unique morphism T → T 0 of algebraic groups, and vice versa. In fact,
you can view X(?) as a contravariant equivalence of categories between
the category of tori and the category of finitely generated free abelian
groups.
All elements of a torus T are semisimple. So if V is any finite di-
mensional representation of T , every element of T is diagonalizable in
its action on V by the Jordan decomposition. Moreover, they commute,
hence we can actually diagonalize
M
V = Vλ
λ∈X(T )
where
Vλ = {v ∈ V | tv = λ(t)v for all t ∈ T }.
154
16.1 Maximal tori and the root system 155
As before, the Vλ ’s are called the weight spaces of V with respect to the
torus T .
Now let G be an arbitrary connected algebraic group. A maximal
torus of G is what you’d think: a closed subgroup T that is maximal
subject to being a torus. Let me state some theorems about maximal
tori in connected solvable groups. These are proved by induction, though
it is often quite difficult...
Theorem 16.1.1 Let G be a connected solvable group. Then, the set

Gu of all unipotent elements of G is a closed connected normal subgroup
of G. All the maximal tori of G are conjugate, and if T is any one of
them, then G is the semi-direct product of T acting on Gu .
As a consequence, you show that in an arbitrary connected group G,

all its maximal tori are conjugate. Indeed, any maximal torus T of G
is contained in a Borel subgroup B. If T 0 is another maximal torus,
contained in a Borel B 0 , we can conjugate B 0 to B to assume that T 0
is also contained in B. But then T and T 0 are conjugate in B by the
theorem.
Now start to assume that G is a reductive algebraic group. Let T
be a maximal torus. Let g be the Lie algebra of G. We can view g as
a representation of T via the adjoint action. It turns out moreover –
using for the first time that G is reductive – that the zero weight space
of g with respect to T is exactly the Lie algebra t of T itself. So we can
decompose
M
g=t⊕ gα
α∈Φ
where Φ is the set of all 0 6= α ∈ X(T ) such that the T -weight space
gα 6= 0. You can already see the root system emerging... The difference
now however is that the set Φ of roots is a subset of the free abelian
group X(T ). Now using the assumption that G is reductive again, you
show:
(1) Each gα is one dimensional, and α ∈ Φ iff −α ∈ Φ.

(2) The group W = NG (T )/T is a finite group that acts naturally on
X(T ) and permutes the subset Φ ⊆ X(T ).
(3) Let Q be the root lattice, the subgroup of X(T ) generated by Φ, and
let E = R ⊗Z Q. Fix a positive definite inner product on E that is
invariant under the action of W . Then, (E, Φ) is an abstract root
system.
156 The classification of reductive algebraic groups
(4) If we embed T into a Borel subgroup B, we get a choice Φ+ of positive
roots defined by α ∈ Φ+ iff gα ⊂ b. Conversely, any choice Φ+ of
positive roots determines a unique Borel subgroup of G containing T .
We’ve now built out of G a root system (E, Φ), and realized the Weyl
group W explicitly as the quotient group NG (T )/T . Moreover, Φ is a
subset of the character group X(T ) of T .
If G is semisimple, then G is determined up to isomorphism by its
root system (E, Φ) together with the extra information given by the
fundamental group X(T )/Q. In the next section, we will see a more
natural setup which classifies the reductive, not just semisimple, groups.
This is harder, since X(T ) will in general be of bigger rank than Q, and
so there is much more freedom not captured by the fundamental group
alone. For GLn , X(T ) is a free abelian group of rank n, whereas Q is of
rank (n − 1).
16.2 Sketch of the classification

Finally let’s prepare the way to state the classification of reductive al-
gebraic groups in general. Let G be a reductive algebraic group, and let
T be a maximal torus. Let Φ ⊂ X(T ) be the root system of G, defined
from the decomposition
M
g=t⊕ gα .
α∈Φ
Let
X(T ) = Hom(T, Gm )
be the character group of T , and let
Y (T ) = Hom(Gm , T )
be the cocharacter group. This is also a free abelian group of rank dim T .
Moreover, there is a pairing
X(T ) × Y (T ) → Z
defined as follows. Given λ ∈ X(T ) and ϕ ∈ Y (T ), the composite λ ◦ ϕ

is a map Gm → Gm . So since Aut(Gm ) = Z,
(λ ◦ ϕ)(x) = xhλ,ϕi
for a unique hλ, ϕi ∈ Z.

For each α ∈ Φ, you prove that there is a (unique up to scalars)
homomorphism
xα : Ga → G
such that
txα (c)t−1 = xα (α(t)c)
for all c ∈ Ga , t ∈ T , such that the tangent map
dxα : L(Ga ) → gα
is an isomorphism. Moreover, the xα ’s can be normalized so that there
is a homomorphism
ϕα : SL2 → G
such that

1 c 1 0
ϕα = xα (c), ϕα = x−α (c).
0 1 c 1
Define

∨ ∨ c 0
α : Gm → T, α (c) = ϕα .
0 c−1
So α∨ ∈ Y (T ). This is called the coroot associated to the root α ∈ Φ.
Now we have built a datum (X(T ), Φ, Y (T ), Φ∨ ), where Φ∨ is the set
of all coroots. This is the root datum of G with respect to the torus T .
(Actually, since all maximal tori in G are conjugate, it doesn’t depend
up to isomorphism on the choice of T .) The notion of root datum is
the appropriate generalization of root system to take care of arbitrary
reductive algebraic groups, not just the semisimple ones.
Here is an axiomatic formulation of the notion of root datum: a root
datum is a quadruple (X, Φ, Y, Φ∨ ) where
(a) X (“characters”) and Y (“cocharacters”) are free abelian groups of

finite rank, in duality by a pairing h., .i : X × Y → Z;
(b) Φ ⊂ X (“roots”) and Φ∨ ⊂ Y (“coroots”) are finite subsets, and there
is a given bijection α 7→ α∨ from Φ to Φ∨ .
To record the additional axioms, define for α ∈ Φ the endomorphisms
sα , s∨
α of X, Y respectively by
sα (x) = x − hx, α∨ iα, s∨ ∨

α (y) = y − hα, yiα .
Then we have the axioms:

158 The classification of reductive algebraic groups
(RD1) For α ∈ Φ, hα, α∨ i = 2.
(RD2) For α ∈ Φ, sα Φ = Φ, s∨ ∨ ∨
αΦ = Φ .
The datum (X(T ), Φ, Y (T ), Φ∨ ) built from our algebraic group G earlier

is such a gadget.
There is a notion of morphism of root datum
(X, Φ, Y, Φ∨ ) → (X 0 , Φ0 , Y 0 , (Φ0 )∨ ) :
a map f : X 0 → X that maps Φ0 bijectively onto Φ and such that the
dual map f ∨ : Y → Y 0 maps f (α)∨ to α∨ for all α ∈ Φ0 . Hence there is
a notion of isomorphism of root datums.
Now suppose that G, G0 are reductive algebraic groups with maximal
tori T, T 0 respectively and corresponding root data (X(T ), Φ, Y (T ), Φ∨ )
and the primed version. Let f : (X(T ), . . . ) → (X 0 (T ), . . . ) be a mor-
phism of root data. It induces a dual map f : T → T 0 of tori. The step
is to show that f can be extended to a homomorphism f¯ : G → G0 .
Using it you prove in particular the isomorphism theorem:
Theorem 16.2.1 Two reductive algebraic groups G, G0 are isomorphic

if and only if their root datums (relative to some maximal tori) are
isomorphic.
There is also an existence theorem:
Theorem 16.2.2 For every root datum, there exists a corresponding

reductive algebraic group G.
Finally, one intriguing thing: given a root datum (X, Φ, Y, Φ∨ ) there
is the dual root datum (Y, Φ∨ , X, Φ). If G is a reductive algebraic group
with root datum (X, Φ, Y, Φ∨ ) you see there is a dual group G∨ with the
corresponding dual root datum. Note the process of going from G to
G∨ is very clumsy: I don’t think there is any direct way of constructing
the dual group out of the original.
Example 16.2.3 Suppose that G is a semisimple algebraic group. Let

Q = ZΦ ⊂ X(T ). Here, Q and X(T ) have the same rank, so Q is a lattice
in X(T ), and X(T )/Q is a finite group, the fundamental group. Let P
be the dual lattice to Q. Fixing a positive definite W -invariant inner
product on E = R ⊗Z Q, we can identify P with the weight lattice of the
root system of G, and then everything is determined by the relationship
between Q ⊆ X(T ) ⊆ P . You can formulate the classification just of
the semisimple algebraic groups in these terms.
Example 16.2.4 Let G be a semisimple algebraic group, and suppose
that Q ⊆ X(T ) ⊆ P are as in the previous example. If X(T ) = P , then
G is called the simply-connected group of type Φ. If X(T ) = Q, then
G is called the adjoint group of this type. Now let Gsc be the simply-
connected one, Gad be the adjoint one. Let G be any other semisimple
group of type Φ. Then, there is an inclusion X(T ) ,→ P = X(Tsc ). This
induces a map Gsc G. Similarly, there is always a map G Gad .
Example 16.2.5
(1) Consider the root datum of GL2 . Here, X(T ) has basis ε1 , ε2 ,
these being the characters picking out the diagonal entries. More-
over, the positive root is α = ε1 − ε2 . Also Y (T ) has basis ε∨ ∨
1 , ε2 ,
the dual basis, mapping Gm into each of the diagonal slots. The
coroot is α∨ = ε∨ ∨
1 − ε2 .
(2) GL2 is its own dual group.
(3) Consider the root datum of SL2 × Gm . Here, X(T ) has basis
α/2, ε, Y (T ) has the dual basis α∨ , ε∨ (here α is the usual positive
root of SL2 ).
(4) Consider the root datum of P SL2 × Gm . Here, X(T ) has basis
α, ε, Y (T ) has the dual basis α∨ /2, ε. So P SL2 × Gm is the dual
group to SL2 × Gm .
(5) As an exercise in applying the classification, you can show that
(1),(3) and (4) plus one more, the 4 dimensional torus, are all the
reductive algebraic groups of dimension 4.
Example 16.2.6 Here are some more examples of dual groups (I think!).
The dual group to SLn is P SLn . The dual group to Sp2n is SO2n+1 .
The dual group to P Sp2n is Spin2n+1 . The dual group to SO2n is SO2n .
The dual group to Spin2n is P SO2n .
For more explicit constructions of root datums, see Springer, 7.4.7.
Bibliography
[Br] J. Brundan, Lecture Notes on Algebraic Groups.

[Hu] J. Humphreys, Linear Algebraic Groups.
[Ma] H. Matsumura, Commutative Algebra.
[Sh] I.R. Shafarevich, Basic Algebraic Geometry.
[Sp] T.A. Springer, Linear Algebraic Groups.
[St] R. Steinberg, Lectures on Chevalley Groups.
160

Lectures On Algebraic Groups: Alexander Kleshchev

Загружено:

Сведения о документе

Оригинальное название

Авторское право

Доступные форматы

Поделиться этим документом

Поделиться или встроить документ

Параметры публикации

Этот документ был вам полезен?

Это неприемлемый материал?

Авторское право:

Доступные форматы

Lectures On Algebraic Groups: Alexander Kleshchev

Загружено:

Авторское право:

Доступные форматы

Lectures on Algebraic Groups

Part one: Algebraic Geometry page 1

Definition 1.0.1 A functor F : A → B is called faithful if the map

HomA (A1 , A2 ) → HomB (F (A1 ), F (A2 )), θ 7→ F (θ) (1.1)

is injective, and F is called full if the map (1.1) is surjective.

Theorem 1.0.2 A functor F : A → B is an equivalence of categories if

Proof ( ⇒ ) Let F be an equivalence of categories and G : B → A

Hence ϕ can be recovered from F (ϕ) by the formula

ϕ = αA2 ◦ GF (ϕ) ◦ (αA1 )−1 . (1.2)

This shows that F is faithful. Similarly, G is faithful. To prove that F

2.1 Some random facts

Lemma 2.1.1 Let k be a field, f, g ∈ k[x, y], and assume that f is

Proof See [Sh, 1.1].

Proposition 2.1.2 Let A, B be k-algebras, I / A, J / B be ideals. Then

A/I ⊗k B/J → (A ⊗k B)/(A ⊗ J + I ⊗ B), ā ⊗ b̄ 7→ a ⊗ b

Definition 2.1.3 A subset S of a commutative ring R is called multi-

Lemma 2.1.4 Let S ⊂ R be a proper multiplicative set. Let I be an

Theorem 2.1.5 (Prime Avoidance Theorem) Let P1 , . . . , Pn be

Proof We can assume that none of the prime ideals is contained in

Lemma 2.1.6 (Nakayama’s Lemma) Let M be a finitely generated

Proof Let m1 , . . . , ml be generators of M . The condition IM = M

for some xij ∈ I. Hence

So by Cramer’s rule, dmj = 0, where d = det(xij − δij ). Hence dM = 0.

Corollary 2.1.7 If B ⊃ A is a ring extension, and B is finitely gener-

Proof Since B contains 1, we have aB = 0 only if a = 0. Now all

Proof Apply Lemma 2.1.6 to M/M 0 .

Corollary 2.1.9 (Nakayama’s Lemma) Let M be a finitely generated

Proof Localize at I and apply Corollary 2.1.8.

Corollary 2.1.10 Let M be a finitely generated module over the ring A

Proof Apply Corollary 2.1.8 to M 0 = (m1 , . . . , mn ).

Lemma 2.1.11 Let M be a maximal ideal of R, then the map R →

A field extension K/k is called separable, if either char k = 0 or

Let L/E be a field extension. A derivation is a map δ : E → L such

δ(x + y) = δ(x) + δ(y) and δ(xy) = xδ(y) + δ(x)y (x, y ∈ E).

If F is a subfield of E, then the derivation δ is F -derivation if it is F -

dim DerF (E, L) = tr. degF E.

(ii) E/F is separable if and only all derivations F → L extend to

Theorem 2.1.15 [Ma, Theorem 20.3] A regular local ring is a UFD. In

2.2 Ring extensions

Definition 2.2.1 A ring extension of a ring R is a ring A of which R is

If A is a ring extension of R, A is a fathful R-module in a natural way.

Definition 2.2.2 A ring extension A of R is called finitely generated if

The following notion resembles that of an algebraic element for field

In Proposition 2.2.5 we give two equivalent reformulations of the in-

Lemma 2.2.4 Let V be an R-module. Assume that v1 , . . . , vn ∈ V and

Proposition 2.2.5 Let A be a ring extension of R and α ∈ A. The

Proof (i) ⇒ (ii) Assume f (α) = 0 , where f (x) ∈ R[x] is monic of

αvi = ai1 v1 + · · · + ain vn (1 ≤ i ≤ n).

As v1 , . . . , vn generate V , this implies that D annihilates V . As V is

Lemma 2.2.6 Let R ⊆ A ⊆ B be ring extensions. If A is finitely

Proof If a1 , . . . , am are generators of the R-module A and b1 , . . . , bn

Proposition 2.2.7 Let A be a ring extension of R.

Proof (i) Let α ∈ A. Then A is a faithful R[α]-module, and we can

Corollary 2.2.8 Let A be a ring extension of R. The elements of A

Proof If α1 , α2 ∈ A are integral, then α1 − α2 and α1 α2 belong to

Definition 2.2.9 The integral closure of R in A ⊇ R is the ring R̄ of all

Example 2.2.10 The elements of the integral closure of Z in C are

Proposition 2.2.11 Let R, A, B be rings.

Proof (i)-(iii) is an exercise.

Proposition 2.2.12 Every U F D is integrally closed.

Proof Let R be a UFD and F be its field of fractions. Let ab ∈ F be

Proposition 2.2.13 If a domain R is integrally closed, then so is S −1 R

Example 2.2.14 The ring Z[i] of Gaussian integers is Euclidean (the